agnes-the-ai-analyst/app/api/marketplace.py
ZdenekSrotyr b4d3c576af
Activity Center: audit log + telemetry + sessions + agnes_* tables (#278)
* docs(spec): admin observability spec + Activity Center MVP plan

Parent spec (480 lines) + executable plan (2295 lines, 14 TDD tasks).
Covers Activity Center rebuild (/admin/activity), with /admin/sessions
and /admin/feedback deferred to follow-up plans.

Already incorporates reviewer-pass revisions across three angles
(security, production resilience, code architecture):
- _get_db import path corrected to app.auth.dependencies
- Test fixtures aligned with seeded_app / admin_user / get_system_db
- All new audit writes wrapped in try/except + logger.exception
- Filename sanitization on session uploads
- DuckDB DESC index behavior documented; upgrade window flagged
- Migration idempotency + evolved-DB test cases
- reveal_raw + shared-cache multi-worker explicitly deferred

Targets schema v40 (audit_log gains params_before, client_ip,
client_kind, correlation_id + 3 indices).

* feat(db): schema v40 — audit_log gains params_before, client_ip, client_kind, correlation_id + 3 indices

* chore(test): clean up Task 1 — drop unused import, rename stale test

* feat(audit): AuditRepository.log() accepts params_before/client_ip/client_kind/correlation_id

* test(audit): strengthen params_before assertion to round-trip JSON content

* feat(audit): AuditRepository.query() rich filters + keyset cursor pagination

* feat(sync): SyncStateRepository.list_recent() cross-table feed

* feat(audit): POST /api/sync/trigger writes audit_log row

* feat(audit): POST /api/scripts/run-due writes audit_log row

* feat(audit): POST /api/upload/sessions writes audit_log row + sanitizes filename

* feat(audit): GET /api/data/{table_id}/download writes audit_log row

* feat(activity): /api/admin/activity timeline + /health + /sync endpoints

* feat(ui): /admin/activity rebuilt — health pulse, timeline, sync grid; /activity-center → 308 redirect

BREAKING: removed demo executive-pulse / maturity-roadmap content from activity_center.html.
The page now reflects real audit_log + sync_history data.

* feat(ui): admin nav + dashboard widget point at /admin/activity

* feat(activity): recursive-audit suppression for AC read endpoints (60s window per actor+filter)

* feat(activity): emit PostHog events when integration enabled (no-op default)

* fix(audit): move v40 indices out of _SYSTEM_SCHEMA + update test_repositories to unpack query() tuple

_SYSTEM_SCHEMA CREATE INDEX on audit_log(timestamp) failed when migration
tests hand-roll a bare audit_log (id, action) without the timestamp column.
Fix: remove indices from _SYSTEM_SCHEMA; add ADD COLUMN IF NOT EXISTS guards
for timestamp and other pre-v40 columns in _v39_to_v40() so the upgrade path
is safe on any hand-rolled schema; call _v39_to_v40 explicitly in the
fresh-install (current==0) path to restore index creation there.

Also unpack the (rows, next_cursor) tuple from AuditRepository.query() in
the three TestAuditRepository tests that still treated it as a list.

* docs: CHANGELOG entry for Activity Center MVP

* chore: refresh stale module docstring in app/api/activity.py

* feat(cli): agnes admin activity — terminal access to Activity Center (timeline + health + sync)

* fix(db): _v39_to_v40 — add IF NOT EXISTS guard for 'action' column

The v39→v40 ladder step adds defensive ADD COLUMN IF NOT EXISTS for
every audit_log column so a hand-rolled bare audit_log (id only) is
safe through the ladder. 'action' was missing from the guard list,
causing CREATE INDEX idx_audit_action_time to fail on tests that
stub audit_log with only an id column (tests/test_e2e_extract.py::
TestSchemaMigration::test_migration_preserves_and_extends).

Local 6/6 schema tests + the previously-failing CI test pass.

* docs(spec): platform telemetry epic — Boss directive + Activity Monitoring plan rebased onto v40 (stacked on zs/spec-activity-center)

* feat(db): schema v41 — 7 usage_* tables for telemetry (events, summary, rollups, attribution)

* chore(db): tighten v41 — usage_session_summary.session_id NOT NULL + upgrade test asserts all 7 tables

* feat(usage): UsageAttributionRepository — replace/delete/lookup over usage_attribution_* tables

* refactor(marketplace): extract list_inner_skills/agents/commands to src/marketplace_listing.py for reuse

* feat(usage): explode plugin attribution on marketplace sync + store entity write; backfill script

* refactor(marketplace): finish src/marketplace_listing.py extraction — drop duplicate _list_inner_* + _parse_frontmatter from app/api/marketplace.py

* feat(usage): promote attribution helpers to src/usage_attribution_helpers.py; hook update_entity rename + bundle-swap; clarify best-effort semantics

* feat(usage): UsageProcessor real extraction + rollup rebuild + 10 fixture-driven tests

* fix(usage): include tool_id in event hash + executemany + rollup transaction (critical multi-tool-turn drop fix)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(marketplace): popularity stats — invocations_30d + trend + sort=most_used|trending + Most Popular section

* feat(admin): /admin/users/<id> Sessions section — list + single-file + bulk-zip downloads (audit-logged)

* feat(usage): admin export endpoint + CLI — csv/json/parquet streaming, filters, audit-logged

* feat(usage): agnes admin ask — LLM Text-to-SQL over usage_events with SELECT-only validator (audit-logged)

* feat(usage): reprocess + prune endpoints + scheduler daily prune job + CLI

* docs: PLATFORM_SETUP.md operator playbook + HOWTO/ cookbook (5 guides + index)

Adds docs/PLATFORM_SETUP.md as a consolidated operator playbook covering
bootstrap, TLS, marketplaces (curated + flea), scheduler env vars, telemetry
extraction/export/ask/prune, privacy posture, and daily routine.

Adds docs/HOWTO/ with 5 analyst cookbook guides: first query, snapshots for
remote tables, private sessions, feedback + admin ask, and customizing skills.

Existing setup docs (QUICKSTART, DEPLOYMENT, ONBOARDING, HEADLESS_USAGE)
get a one-line cross-reference at the top pointing to PLATFORM_SETUP.md.

* docs(changelog): platform telemetry epic — usage_* foundation + surfaces + admin access + docs

Comprehensive [Unreleased] entry covering: usage_events/session_summary/
tool_daily/plugin_daily tables (v41), attribution lookup tables, backfill
script, marketplace Most Popular + invocation chips + sort, admin Sessions
section, export/ask/reprocess/prune endpoints + CLI mirrors, Activity Center
(v40), PLATFORM_SETUP.md + HOWTO/ docs, and operations notes for v41 upgrade.

* fix(security): block DuckDB read_*/http_*/glob functions in usage_ask validator + symlink escape guard in session zip + clarify mark-private semantics

* fix(admin): parquet export tempfile cleanup on COPY failure + correct processed-first sort on /admin/users/<id>/sessions

* feat(audit): close 8 production audit gaps — query (local/remote/hybrid), catalog/schema/sample, snapshot estimate/create, check-access

* feat(ui): /admin/usage summary dashboard + per-user activity tab on /admin/users/<id>

* fix(audit): cap error messages at 200 chars + audit user_activity reads + recursion guard on usage.summary

* fix(audit): catalog.list audits on error path + clean up deferred json import

* fix(ux): client_kind=cli for PAT auth + timeline empty state + email-instead-of-uuid + nav reorder + help text + loading indicators + ask doc

* feat(observability): unify /admin/activity into single page with saved views

- KPI cards (events, users, error rate, p95) clickable as quick-filters
- Faceted filter dropdowns populated from audit_log in the current window
- Sortable audit table, cursor pagination, per-row JSON side panel
- Saved views (schema v43: user_observability_views) — per-user state
- Top bar: window selector + 30s Live toggle + saved views dropdown
- /admin/scheduler-runs → 308 redirect (source=scheduler filter)
- New endpoints: /api/admin/observability/{facets,kpis,views}

* test: update activity + scheduler-runs tests for unified page

- test_admin_activity_page_renders asserts new structural anchors
- test_admin_scheduler_runs_page_admin_only asserts 308 redirect

* fix(observability): respect [hidden] on modal + side panel

CSS `display: flex` on .obs-modal beat the [hidden] attribute's UA
display:none, so the save-view modal rendered on page load and Cancel
clicks couldn't dismiss it. Gate the modal's flex layout on
:not([hidden]); add the same display:none guard prophylactically to
.obs-panel and .obs-views-panel.

* feat(observability): user enrichment in audit + interactive /admin/usage

Activity:
- /api/admin/activity now joins users for user_email + user_name per row
- User column renders "name (id-prefix)" or "email (id-prefix)" instead
  of an opaque truncated UUID; falls back to id when the user record is
  missing

Usage:
- /admin/usage rewritten as the same filter/group-by/search pattern as
  /admin/activity. Faceted dropdowns (User / Tool / Source / Event type)
  populated from usage_events; debounced free-text search across
  tool_name / skill_name / subagent_type / command_name
- New endpoints /api/admin/usage/{facets,kpis,query}; the query endpoint
  supports group_by in {day, username, tool_name, source, ref_id} with
  sort + offset pagination, plus an ungrouped raw-events mode
- 4 KPI cards (events, distinct users, distinct tools, error rate) are
  clickable quick-filters; clicking a grouped row applies the bucket as
  a filter
- Old static `?window=7d|30d|all` server preload removed; all state is
  client-side via since_minutes + group_by + filters in the URL

* fix(observability): clearer labels, all-column sort, drop saved views UI

- Rename page titles: "Activity" → "Server activity", "Usage" → "Tool usage"
  with a one-line subtitle on each explaining what the page covers and
  linking the other one. The two pages source different data (audit_log
  vs usage_events) and the previous labels conflated them.
- Drop the saved-views dropdown + save modal from /admin/activity. The
  modal pop-open bug was the trigger; the value wasn't there yet. The
  /api/admin/observability/views CRUD + DuckDB table stay in place.
- Rename "Live (30s)" to "Auto-refresh (30s)" with a tooltip clarifying
  that it's the re-fetch rate, not the time range. Time range now
  labeled "Time range" instead of "Window".
- All audit-table columns are sortable (User, Source, Action, Resource,
  Result added); sort is page-local with a Jinja comment explaining the
  trade-off. Same for raw usage rows.
- Fix duplicate sort-arrow bug — the literal "▼" in the Time th HTML was
  rendering alongside the CSS ::before arrow. Removed the literal; CSS
  is the single source of truth.

* feat(observability): global Sessions browser + transcript viewer + CLI

Web:
- /admin/sessions — list every collected session JSONL across all users
  with time-range, user, model, errors-only and free-text filters. Default
  sort surfaces error-heavy sessions first. KPI cards (sessions, distinct
  users, sessions w/ errors, tool error rate) clickable as quick-filters.
- /admin/sessions/<username>/<file> — transcript viewer rendering the
  JSONL chronologically: user prompts, assistant text, tool calls (with
  JSON input) and tool results (with flattened output). Errors get a red
  border + chip and a "Next error" navigation button at the top.
- Admin dropdown gains a "Sessions" link.

API:
- GET /api/admin/sessions/{list,kpis,facets} — filtered cross-user reads
  off usage_session_summary
- GET /api/admin/sessions/{username}/{file}/transcript — parses JSONL via
  the existing services.session_pipeline.lib, returns chronological events
- GET /api/admin/sessions/{username}/{file}/download — JSONL stream, same
  path-safety guards as the per-user endpoint, audit-logged

CLI:
- `agnes admin sessions list [--user X] [--errors] [--since 7d]` — table
  output with `!` prefix on rows that hit a tool error
- `agnes admin sessions show <username> <file>` — transcript dump, with
  `--errors` to print only the failed tool_result blocks
- `agnes admin sessions download <username> <file> [-o path]`
- `agnes admin sessions kpis` — top-level numbers

* feat(internal): expose telemetry tables to agnes query with row-level RBAC

Three new registered tables backed by system.duckdb, queryable through
the same /api/query plumbing analysts use for Keboola / BigQuery /
local sources:

  agnes_sessions  → usage_session_summary  (filter: username)
  agnes_usage     → usage_events           (filter: username)
  agnes_audit     → audit_log              (filter: user_id)

RBAC is per-row, not per-table: admins see every user's rows; non-admins
see only their own. The filter is built server-side from the auth user
dict; non-admin filter values are regex-validated before SQL interpolation.

Implementation:
- new connector connectors/internal/ with access (filter+exec) + registry
  (idempotent table_registry seed at startup)
- /api/query detects internal table refs and short-circuits to a CTE
  wrapper that prepends "WITH agnes_x AS (SELECT * FROM <src> WHERE …),
  …" then "SELECT * FROM (<user_sql>) AS _q". DuckDB cursor on the
  shared system.duckdb handle — opening parallel handles / ATTACH on the
  same file is blocked process-wide.
- mixing internal + BQ / registered local tables in one SELECT is
  rejected (v1 limitation)
- src.rbac.can_access_table waves internal tables through for all
  authenticated users; row scoping is the actual security control
- /api/v2/schema and /api/v2/sample gained internal branches; sample
  intentionally skips its cache because rows are RBAC-scoped per caller
- audit row written as action='query.internal' with is_admin flag

Tests: connectors/internal/access — RBAC, filter clause, schema, CTE
wrapper coexistence with user-supplied aggregations, unsafe-username
rejection. 16/16 passing.

Motivating queries this enables:
  SELECT tool_name, COUNT(*) FROM agnes_usage
   WHERE is_error GROUP BY 1 ORDER BY 2 DESC
   -- analyst self-introspection: which tools fail for me?

  SELECT user_id, COUNT(*) FROM agnes_audit
   WHERE action = 'session.transcript_view' GROUP BY 1
   -- admin: who's been looking at whose session transcripts?

* feat(admin): group dropdown into 5 named sections + internal tables in /catalog

Admin dropdown gains section headers so admins can land on the right
page without re-reading the full menu:

  Activity Center      Server activity / Tool usage / Sessions
  Users & Access       Users / Groups / Resource access / Tokens
  Data                 Tables
  Agent Experience     Curated Marketplaces / Flea Submissions /
                       Agent Setup Prompt / Agent Workspace Prompt
  Server               Server config

"Agent Experience" frames the curated content + prompts as one cluster
— it's all admin-controlled material that shapes what an analyst's AI
agent encounters. "Configuration" → "Server" since only one item lives
there now.

Renamed the section's first two items:
  "Activity" → "Server activity" (matches page H1)
  "Usage"    → "Tool usage"

Also fixes /catalog visibility of the internal tables (agnes_sessions /
_usage / _audit) for non-admin users: ``app.auth.access.can_access``
short-circuits to True for resource_type='table' + an internal-table id.
Without this, non-admins saw the tables in /api/v2/catalog (which uses
the same RBAC bypass) but not on the /catalog HTML page (which calls
can_access directly, requiring a resource_grants row internal tables
don't have).

CSS for `.app-nav-menu-section`: small caps, muted, non-clickable; first
section trims top padding so the panel doesn't open with an awkward gap.

* refactor(admin): move corporate memory into Admin > Agent Experience

Memory link was the only admin-only entry in the primary nav (gated by
session.user.is_admin). Moves it into the Admin dropdown under Agent
Experience, alongside Curated Marketplaces / Flea Submissions / Prompts
— all admin-curated content that shapes what an analyst's AI agent
encounters.

Renamed the nav label to "Shared Knowledge" to match what the page
actually is (admin-curated organisational knowledge from session
verification, surfaced to agents). URL stays at /corporate-memory; the
route still gates on require_admin per the existing comment.

Side effect: primary nav (Home / Marketplace / Data Packages) is now
uniform for every authenticated user — no conditional admin-only entry.

* ui: rename admin entries to Curated Knowledge / Init Prompt / Workspace Prompt

- "Shared Knowledge" → "Curated Knowledge" (parallel with "Curated
  Marketplaces" in the same Agent Experience section; "curated" tells
  the admin what they do there — review + approve)
- "Agent Setup Prompt" → "Init Prompt" (matches the `agnes init` flow
  it actually drives)
- "Agent Workspace Prompt" → "Workspace Prompt" (the "Agent" prefix
  was redundant — every item in the section is agent-facing)

Renames page titles + H1s on /admin/agent-prompt and
/admin/workspace-prompt to match.

* refactor: rename Usage → Telemetry across user-facing surfaces

External surfaces all switch; internal Python module / file names and the
physical DB tables (usage_events, usage_session_summary, usage_tool_daily,
usage_plugin_daily) stay — renaming them would force a schema migration
+ a redo of the LLM Text-to-SQL prompt for no analyst-visible win.

Changes:
- Admin dropdown: "Tool usage" → "Telemetry"
- Page H1 / <title>: same
- URL: /admin/usage → /admin/telemetry; old URL 308-redirects
- API prefix: /api/admin/usage/* → /api/admin/telemetry/*
- CLI: primary command `agnes admin telemetry …`; `agnes admin usage` kept
  as a deprecated alias so existing operator scripts keep working
- Internal data-source table id: agnes_usage → agnes_telemetry. The
  registry seed now evicts any stale internal-source row whose id no
  longer matches INTERNAL_TABLES, so the old `agnes_usage` row is
  removed from table_registry on next app boot
- All tests + JS endpoint paths updated

* test(rbac): include auto-appended internal tables in expectations

get_accessible_tables now appends agnes_sessions / agnes_telemetry /
agnes_audit to every authenticated user's accessible-tables list so the
internal data source shows up in /catalog. The two existing rbac tests
asserted hardcoded list shapes that pre-dated the change.

Rewritten to assert "granted tables + the canonical internal-table set"
instead of literal lists, so the test stays correct if the internal
table roster changes again later.

* ui: visual dividers between admin-dropdown sections

Adds a 1px top border + 6px top margin to every section header except
the first, so the five named groups (Activity Center, Users & Access,
Data, Agent Experience, Server) read as visually separated clusters.
The header itself stays small-caps + muted as before — the border is
additive.

* ui(memory): match obs-topbar visual on /corporate-memory

The Curated Knowledge page (linked from the admin dropdown's Agent
Experience section) opened straight into the stats bar — no title,
no subtitle, no shared chrome with the other admin pages. Adds an
obs-topbar-style header at the top of .container-memory:

  - H1 "Curated Knowledge"
  - subtitle explaining what the page is + how AI agents pull from it

The `.ck-*` class set duplicates the inline obs-* styles from
/admin/activity etc. for this one page; promoting the obs-* class set
to style-custom.css for shared reuse is the obvious next step (4 pages
already inline the same CSS), tracked as a follow-up.

Page <title> also renamed from "Corporate Memory" → "Curated Knowledge".

* ui(tables): list Agnes internal tables in /admin/tables + group in /catalog

/admin/tables previously rendered three per-source-type listings
(BQ / Keboola / Jira) and dropped any row whose source_type didn't
match — so the agnes_sessions / agnes_telemetry / agnes_audit rows
seeded into table_registry were invisible. Adds a fourth read-only
section "Agnes internal tables" that filters source_type === 'internal'
and renders the same registry-table layout the other sections use,
with two changes:

  - no Register button (these rows are seeded on every app boot from
    connectors/internal/registry.py)
  - Edit + Delete actions hidden (any change would be reverted on the
    next start). Manage access stays so admins can still inspect.

Mode badge picks up a new mode-internal CSS class (teal accent) so the
display doesn't lie and call it "local".

In /catalog, internal tables now group under an "agnes" accordion
section (bucket="agnes" on seed) instead of falling into the catch-all
"default". Single source of truth for which tables exist; admins find
them where they expect.

* ui(tables): Agnes internal as a 4th tab next to BQ/Keboola/Jira

Previous iteration mounted the internal-table listing as a separate
standalone card under the tab strip. Reshapes it to a proper
tab-content section so admins switch between data sources via one
consistent nav (BigQuery / Keboola / Jira / Agnes internal).

- New tab button "Agnes internal" in the tab-nav.
- The listing card becomes <section id="tab-content-internal"
  class="tab-content">; switchTab() already routes by id so no JS
  change beyond extending the hash allowlist for direct #internal
  links.
- Tab content keeps the read-only treatment from the previous commit
  (no Register button, no Edit / Delete in renderRegistryListing).

* ui: rename Curated Knowledge → Curated Memory

Settles the naming back on "Curated Memory" — parallel structure with
"Curated Marketplaces" in the same Agent Experience section, and zero
rename ripple: URL (/corporate-memory), API (/api/memory/*), CLI
(agnes admin memory), and Python modules all stay on "memory" so the
admin label finally lines up with the underlying surfaces.

The "Curated" prefix still tells admins what they do on the page
(review pending → approve / mandate / reject) and reads as a sibling
of "Curated Marketplaces" right next to it in the dropdown.

Touches: admin dropdown label, page <title>, page H1. DB tables stay
on knowledge_* (already the canonical naming for the data shape).

* ui: rename "Server activity" → "Audit log"

"Audit log" is what the page actually is — server-side audit_log table
rendered with KPI cards + filter bar + sortable table. The "Server
activity" label confused the term with Claude Code session telemetry
(Telemetry page) and didn't make the source/concept clear.

Touches:
- Admin dropdown nav label
- /admin/activity page H1 + subtitle
- /admin/telemetry subtitle cross-link
- test_activity_api page-renders assertion

URL (/admin/activity) and API (/api/admin/activity/*) stay — the
"activity" name has stuck at the route layer for a year; rerouting
those would churn dashboards/bookmarks for zero analyst-visible win.

* ui(admin-nav): gray band on each section header for clearer separation

Previous iteration used a 1px top border between section labels — the
labels still blended into the items above/below at a glance. Switches
to a light gray background band per section header, extended edge-to-
edge inside the panel via negative horizontal margins. Bolder
font-weight (700) reinforces the separation; bumping the font color
isn't needed because the band itself does the work.

First section's header tucks into the panel's top border-radius so the
band reaches the corners without a gap.

* ui(catalog): rename internal-table category to "Agnes Internal"

`bucket` is what /catalog renders as the accordion category header
verbatim — "agnes" lowercase didn't read as a real category name and
got confused with a system identifier. Bumps to "Agnes Internal".
Seed re-applies on every app boot so existing rows pick up the new
bucket value via `ON CONFLICT (id) DO UPDATE`.

* ui(catalog): split Agnes Internal into its own card on /catalog

Previously the three internal tables landed inside the "Core Business
Data" card under an "Agnes Internal" accordion alongside Keboola / BQ
buckets — readers conflated system telemetry with business datasets,
and the data_stats header counter ("3 tables · ~X rows total") only
ever counted synced rows so internal tables looked invisible.

Split the catalog page into two cards:
- Core Business Data: only non-internal source_types (Keboola, BQ,
  Jira). Accordions group by bucket as before. Stats counter reflects
  this card's tables.
- Agnes Internal: a dedicated card with its own visual treatment
  (teal accent matching the mode-internal badge in /admin/tables).
  Flat list (no accordion — only 3 rows, never grows here), each
  row carries the canonical `agnes query` snippet. Read-only — no
  profiler click, no In-stack toggle, no sync metadata.

Route adds `internal_card` context object; template renders the new
card only when it's non-None.

* fix(rbac): hide internal tables from /admin/access + drop "my" framing

Two related cleanups for the Agnes-internal tables:

1. /admin/access (resource grants) no longer lists them. The
   `can_access` check has a hardcoded internal-table bypass — security
   is row-level (per-request view filter), so a table-grain
   `resource_grants` row would do nothing. Surfacing them in the UI
   let admins set up grants that silently no-op. Filter at the
   `_table_blocks` projection so the UI tree never sees them.

2. Display names drop the analyst-perspective "my" framing:
     "Agnes — my sessions"          → "Agnes sessions"
     "Agnes — my telemetry events"  → "Agnes telemetry events"
     "Agnes — my audit log"         → "Agnes audit log"
   The "my" only makes sense from the querying analyst's seat
   (`SELECT … FROM agnes_sessions` returns *their* rows); on /admin/*
   pages where admin sees / configures them across users, the
   pronoun was misleading. Description text now spells out the
   row-level RBAC contract explicitly.

Display names update via TableRegistryRepository.register's ON CONFLICT
UPDATE on next app boot; no manual cleanup needed.

* ui: subtitle notes about agnes_* tables on each Activity Center page

The recursive observability story — Agnes serves its own audit /
telemetry / session data through the same `agnes query` plumbing
analysts use for business data — wasn't surfaced anywhere on the
admin pages that show that data. Three pages get a one-liner with
the canonical `agnes query` snippet + the RBAC contract (analysts
see their own rows, admin sees all):

- /admin/activity (Audit log)   → agnes_audit
- /admin/telemetry (Tool usage) → agnes_telemetry
- /admin/sessions               → agnes_sessions

Sets up the discovery moment for admins: they're reading the page,
they see "you can query this from Claude Code", they remember it
when an analyst asks "how do I find my own failed tool calls?".

* ui(tables): explain "Show log" empty-state on /admin/tables

Cache warmup log <pre> renders with a dark background and is only
populated by the SSE stream during a Re-warm all run. Opening the
page cold + clicking Show log just revealed a black bar with no
context — admins couldn't tell what they were looking at.

Adds an inline paragraph above the <pre> explaining what the log is,
the row format, when it fills in, and where to find the historical
audit trail (/admin/activity). The actual <pre> stays empty until
SSE events arrive, but the surrounding copy carries the meaning.

* ui(tables): auto-open cache-warmup log on Re-warm all click

A Re-warm all run takes ~24s per remote BQ row. With the <details>
collapsed by default, operators saw the button disable, watched a
quiet ~24s pass, and assumed nothing had happened — the streaming
log was hidden behind a closed disclosure.

Two small JS tweaks:
- cacheWarmupRun() opens the details on click, so streamed lines
  appear without an extra interaction
- cacheWarmupOnStart() hides the inline hint paragraph the moment
  real log content lands, so the dark log block isn't competing
  with redundant context

Hint paragraph also clarifies that only `query_mode='remote'` BQ
rows are warmed — operators with only materialized/internal tables
would see total=0 and the page would "do nothing" by spec.

* ui: trim Agnes internal copy across surfaces

Descriptions had grown to explain the extraction pipeline ("parsed
out of session JSONLs"), the underlying table ("Backed by
usage_session_summary"), the RBAC mechanic ("row-level RBAC at query
time — analysts see their own; admin sees all"), and the SQL snippet.
Every implementation detail meant another rewrite on the next iter.

Strips to one stable line per surface: what the data is, plus
"Also available locally for analysis". Mechanics live in code +
docs; the page copy says what the user needs to know.

Touched:
- connectors/internal/access.py: INTERNAL_TABLES descriptions
- activity_center.html / admin_usage.html / admin_sessions.html
  subtitles
- catalog.html Agnes Internal card description + row strip
- admin_tables.html "Agnes internal" tab hint

* fix(internal): is_user_admin arity bugs + + saved-view payload cap

Round-1 code review (PR #278) caught two blocking bugs and three nits.

Blocking — both `is_user_admin(user)` (single dict arg) calls raised
TypeError. is_user_admin signature is `(user_id, conn)`. Affected:

- app/api/query.py:_run_internal_query — every POST /api/query that
  references agnes_sessions / agnes_telemetry / agnes_audit blew up
  with a 500. The headline analyst-facing feature of this PR was
  unusable through the API.
- app/api/v2_sample.py — same shape; `GET /api/v2/sample/agnes_*`
  returned 500.

Both fixed to call `is_user_admin(user.get("id"), conn)`. Added two
FastAPI-level tests in test_internal_data_source.py that go through
the TestClient — the existing unit tests on `execute_internal_query`
and `build_filter_clause` skipped the request-handler layer where the
bugs lived, which is why this landed.

Nits also closed:
- connectors/internal/access.py: `+` allowed in _USERNAME_RE /
  _USER_ID_RE so RFC 5321 email local-parts (alice+test@x) resolve
  correctly without hitting InternalAccessError.
- app/api/observability.py: saved-view payload capped at 64 KiB to
  prevent an admin from bloating system.duckdb with a malformed save.

* fix(security): close non-admin data-leak via underlying-table refs

PR #278 R2 review surfaced a non-admin-exploitable bypass: SQL whose
string literal contains 'agnes_sessions' routed into the privileged
internal-query path, then queried the underlying physical table
(usage_session_summary / usage_events / audit_log) directly, escaping
the CTE wrapper's row filter. Two reinforcing defenses:

1. find_internal_refs() now strips single-quoted string literals
   before scanning for alias names — a literal alone no longer
   routes the request into the privileged code path.

2. execute_internal_query() rejects non-admin SQL that references
   the underlying physical tables (usage_*, audit_log). The CTE
   wrapper only scopes the agnes_* aliases; a direct FROM on the
   base table — or a shadowing inner WITH that still has to read
   the base table — bypasses RBAC. Block before execution with an
   actionable error pointing to the agnes_* alias. Admins are
   unaffected (god-mode short-circuit on the filter clause).

3. tests/test_internal_data_source.py — three new negative tests
   covering literal-only matches, direct-table refs, and CTE
   shadow attempts.

Also tightens usage_ask.py's SELECT-only validator: pragma_table_info,
pragma_storage_info, pragma_database_*, and duckdb_tables / columns /
views / indexes / schemas are reflection functions that leak metadata
the analyst question shouldn't reach. \bPRAGMA\b in _FORBIDDEN never
matched the function-call form (word-boundary between `A` and `_`).

* fix(security): dynamic denylist for non-admin internal queries

R3 review (PR #278) caught a wider data-leak than R2: the underlying-
physical-table guard listed only the 7 usage_* + audit_log tables,
but system.duckdb has 30+ other sensitive tables — users (emails +
ids), personal_access_tokens, resource_grants, user_groups,
user_observability_views, store_*, marketplace_*, knowledge_*, etc.
A non-admin SQL like

    SELECT * FROM agnes_sessions
    UNION ALL SELECT email, id, … FROM users LIMIT 1

would leak every user's row.

Replaces the hardcoded denylist with a **dynamic allowlist** —
non-admin SQL may reference ONLY the registered agnes_* aliases.
Every other table in `information_schema.tables` (main schema) is
rejected. Future migrations that add a new sensitive table are
automatically covered without re-editing this module.

Also strips SQL comments (`/* */` and `--`) before the identifier
scan so a comment-wrapped table name (`/**/users/**/`) can't slip
past the regex.

Four new negative tests pin: `users`, `personal_access_tokens`,
block-comment wrap, line-comment wrap.

Plus: per-user view-count cap (100) on /api/admin/observability/views
so an admin can't fill system.duckdb with thousands of saved views.

* release: 0.54.0 — Activity Center + Telemetry + Sessions + internal datasource

Cuts the work shipped across this PR (Activity Center build, recursive
internal data source) into a versioned release. Bumps pyproject.toml
to 0.54.0; renames the top of CHANGELOG.md from [Unreleased] to
[0.54.0] — 2026-05-12 with a header summary; opens a fresh
[Unreleased] section for the next round.

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-12 22:41:19 +02:00

2218 lines
86 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""Unified Marketplace API — backs the ``/marketplace`` browse page.
Combines two upstream sources into a single browse + detail surface:
* **curated** — admin-registered git marketplaces, RBAC-scoped via
``resource_grants`` (resource_type='marketplace_plugin').
* **flea** — community Store uploads (``store_entities``).
Per-tab listing endpoints (``/items``, ``/categories``) plus curated-side
detail and explicit-install (Model B) endpoints. Flea-side install/uninstall
keeps using the existing ``/api/store/entities/{id}/install`` POST/DELETE
endpoints — no need to duplicate.
Curated detail / install endpoints are gated by
``require_resource_access(MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}")``
so a user without the RBAC grant gets a 403 even on direct URL hit.
"""
from __future__ import annotations
import logging
import re
from collections import OrderedDict
from pathlib import Path
from typing import Any, Dict, List, Literal, Optional, Tuple
import duckdb
from fastapi import APIRouter, Depends, HTTPException, Query
from fastapi.responses import FileResponse
from pydantic import BaseModel
from app.auth.access import (
_user_group_ids,
is_user_admin,
require_resource_access,
)
from app.auth.dependencies import _get_db, get_current_user
from app.resource_types import ResourceType
from app.utils import get_marketplace_cache_dir, get_marketplaces_dir
from src.marketplace_filter import (
resolve_allowed_plugins,
resolve_manifest_name,
)
from src.marketplace_listing import _FRONTMATTER_RE, _parse_frontmatter
from src.marketplace_urls import mirrored_url
from src.repositories.audit import AuditRepository
from src.repositories.marketplace_plugins import MarketplacePluginsRepository
from src.repositories.marketplace_registry import MarketplaceRegistryRepository
from src.repositories.store_entities import StoreEntitiesRepository
from src.repositories.user_curated_subscriptions import (
UserCuratedSubscriptionsRepository,
)
from src.repositories.user_store_installs import UserStoreInstallsRepository
from src.store_categories import STORE_CATEGORIES
from src.store_naming import suffixed_name
logger = logging.getLogger(__name__)
router = APIRouter(prefix="/api/marketplace", tags=["marketplace"])
OWNER_TODO_PLACEHOLDER = "owner_todo"
"""Placeholder displayed in the UI when a curated plugin has no owner / curator
metadata. To be replaced once ``marketplace_plugins.curator_owner`` lands."""
# ---------------------------------------------------------------------------
# Pydantic response models
# ---------------------------------------------------------------------------
class MarketplaceItem(BaseModel):
id: str
source: Literal["curated", "flea"]
name: str
type: Literal["skill", "agent", "plugin"]
category: Optional[str] = None
description: Optional[str] = None
owner: Optional[str] = None
version: Optional[str] = None
photo_url: Optional[str] = None
added: Optional[str] = None
installed: bool = False
marketplace_slug: Optional[str] = None
marketplace_name: Optional[str] = None
detail_url: str
# v32+ quarantine UX: surface the card's visibility + an
# is_viewer_owner flag so the listing template can render an
# "Under review" / "Quarantined" corner badge on the submitter's
# own non-approved cards. Approved cards omit both fields.
visibility_status: Optional[str] = None
is_viewer_owner: bool = False
# v39: drives the "Required" pill on the curated browse cards. Only
# set on curated items (flea/store entities are never system).
is_system: bool = False
# Rich-content fields from marketplace-metadata.json (plugin-level only
# for now; skill/agent rich content lands in a later phase). Frontend
# falls back to `name` / `description` when these are missing, so the
# card renders identically to pre-enrichment behaviour for any plugin
# whose curator hasn't filled the new fields yet.
display_name: Optional[str] = None
tagline: Optional[str] = None
# telemetry (Phase B.1): populated from usage_plugin_daily rollup
invocations_30d: int = 0
unique_users_30d: int = 0
trend_pct: Optional[float] = None
class ItemListResponse(BaseModel):
items: List[MarketplaceItem]
total: int
page: int
page_size: int
class CategoryEntry(BaseModel):
name: str
count: int
icon_key: str
class CategoriesResponse(BaseModel):
items: List[CategoryEntry]
class InnerItemSummary(BaseModel):
name: str
description: Optional[str] = None
detail_url: Optional[str] = None # nested detail for skill/agent only
# v32: marketplace-metadata-driven cover photo for the skill/agent card on the
# parent plugin detail page. Already in served-URL form (internal /asset/
# endpoint, mirrored /mirrored/ endpoint, or pass-through external URL).
# None → frontend renders initials placeholder ("SK" / "AG").
cover_photo_url: Optional[str] = None
class CommandEntry(BaseModel):
name: str
description: Optional[str] = None
class HookEntry(BaseModel):
name: str
event: Optional[str] = None # "—" rendered client-side when None
description: Optional[str] = None
class McpEntry(BaseModel):
name: str
type: Optional[str] = None # "stdio" / "sse" / etc.
description: Optional[str] = None
class FileEntry(BaseModel):
"""One file in a plugin / skill / agent bundle. Path is relative to the
bundle root; size is bytes (rendered with ``humanbytes`` on the client)."""
path: str
size: int
class DocEntry(BaseModel):
"""One doc file shipped alongside a Store entity. URL points at the
serving endpoint (``/api/store/entities/<id>/docs/<filename>``)."""
name: str
url: str
class UseCase(BaseModel):
"""One "When to use it" example from marketplace-metadata.json.
Curator-authored: friendly title + 1-2-sentence description + the
literal prompt the user would paste into Claude Code."""
title: str
description: str
prompt: str
class SampleInteraction(BaseModel):
"""Example dialog shown in the "Sample interaction" section.
``assistant`` is markdown — the API pre-renders to safe HTML in
``assistant_html`` so the template can inject without a client-side
markdown lib. ``assistant`` stays as the source for copy-paste / future
re-rendering needs.
"""
user: str
assistant: str
assistant_html: str
class PluginDetailResponse(BaseModel):
"""Unified detail response for a plugin (curated *or* flea).
Frontend-side switching is purely cosmetic — `source` controls the
Curated/Flea pill, photo URL fallback path, and owner label.
"""
source: Literal["curated", "flea"]
# IDs / breadcrumbs
marketplace_id: Optional[str] = None # curated only
marketplace_name: Optional[str] = None # curated only
entity_id: Optional[str] = None # flea only
plugin_name: str
manifest_name: str
# Display
description: Optional[str] = None
version: Optional[str] = None
category: Optional[str] = None
author_name: Optional[str] = None # curated curator / flea owner
curator_email: Optional[str] = None # curated only — surfaced for "contact curator"
owner_display: Optional[str] = None # flea: users.name → email → owner_username
homepage: Optional[str] = None
cover_photo_url: Optional[str] = None # /api/store/.../photo for flea, marketplace-metadata for curated
video_url: Optional[str] = None # v32: external (YouTube/Vimeo/Loom) embed URL
bundle_size: Optional[int] = None # bytes; None when unknown
install_count: int = 0 # flea only; curated leaves at 0
released_at: Optional[str] = None # ISO timestamp
updated_at: Optional[str] = None # ISO timestamp
installed: bool = False
# Internal structure
skills: List[InnerItemSummary] = []
agents: List[InnerItemSummary] = []
commands: List[CommandEntry] = []
hooks: List[HookEntry] = []
mcps: List[McpEntry] = []
# Bundle contents — used by the redesigned skill/agent/plugin detail pages
files: List[FileEntry] = []
docs: List[DocEntry] = []
# v32+ quarantine: surface the entity's visibility so the JS install
# path can refuse to render the install button when non-approved.
# Curated entries omit (always live).
visibility_status: Optional[str] = None
# Latest submission verdict for the linked entity — populated only
# for the owner / admin (the same audiences that see the quarantine
# banner). The banner's auto-refresh JS polls this field so it can
# reload the page when an LLM review lands; visibility alone is
# insufficient because `blocked_llm` keeps the entity at
# `visibility_status='pending'`.
submission_status: Optional[str] = None
# v39: drives the disabled install button on the curated plugin
# detail page. The same flag travels via /api/marketplace/items so
# the browse cards can show a "Required" pill.
is_system: bool = False
# Rich-content fields from marketplace-metadata.json (plugin-level, curated
# only — flea entities don't have a metadata layer). All optional; UI
# sections render only when populated.
#
# `display_name` overrides the technical `manifest_name` on the hero h1
# and window titlebar. `tagline` is the friendly subtitle (replacing
# `description` as the hero summary on listing cards + hero).
#
# `description_long_html` is server-side rendered + sanitized markdown
# (see app/markdown_render.render_safe). It powers the "What it does"
# panel; the raw `description` from marketplace.json is the fallback.
display_name: Optional[str] = None
tagline: Optional[str] = None
description_long_html: Optional[str] = None
use_cases: List[UseCase] = []
sample_interaction: Optional[SampleInteraction] = None
# telemetry (Phase B.1)
telemetry: Optional[Dict[str, Any]] = None
# Legacy alias kept so any unmigrated import continues to resolve. The new
# unified model supersedes the curated-only response shape.
CuratedDetailResponse = PluginDetailResponse
class InnerDetailResponse(BaseModel):
marketplace_id: str
plugin_name: str
kind: Literal["skill", "agent"]
name: str
description: Optional[str] = None
body: str
relpath: str # path inside plugin_dir for diagnostics
# Parent plugin metadata + bundle contents — populated so the redesigned
# skill/agent detail page can render hero badges, sidebar rows, and the
# Files section without a second roundtrip to the parent plugin endpoint.
marketplace_name: str = ""
category: Optional[str] = None
parent_author_name: Optional[str] = None
parent_updated_at: Optional[str] = None
# Parent plugin's `name` from `.claude-plugin/plugin.json` — same value
# the synth marketplace.json uses, so /<manifest_name>:<inner_name> is
# exactly what Claude Code accepts after install.
manifest_name: str = ""
bundle_size: Optional[int] = None
files: List[FileEntry] = []
# v32: per-skill / per-agent enrichment from marketplace-metadata.json sub-tree.
# Read at request time from the cloned working tree (not cached in DB) so
# curators can update one inner asset without paying the cost of a full
# plugin-cache rewrite. Three-typed cover URL layout matches plugin
# detail: internal → /asset/, mirror status not yet checked here so
# external URLs pass through unmirrored (we don't run sync mid-request).
cover_photo_url: Optional[str] = None
video_url: Optional[str] = None
docs: List[DocEntry] = []
# Rich user-facing fields (parity with plugin-level rich content from
# the 2026-05-12 redesign). All optional — UI hides each section when
# the corresponding field is absent.
#
# `category` on the response is curator-override-OR-parent-fallback:
# the API hands back the parent plugin's category when the skill/agent
# didn't opt into its own. The override happens in the inner-detail
# handler before the response leaves the API.
#
# `invocation`: curator-provided literal command string ("…&lt;arg&gt;"
# forms welcome). When absent, the template falls back to its computed
# "<manifest_name>:<inner_name>" string.
display_name: Optional[str] = None
tagline: Optional[str] = None
description_long_html: Optional[str] = None
use_cases: List[UseCase] = []
sample_interaction: Optional[SampleInteraction] = None
when_to_use_html: Optional[str] = None
invocation: Optional[str] = None
class InstallActionResponse(BaseModel):
installed: bool
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def _load_invocation_stats(
conn: duckdb.DuckDBPyConnection,
source: str,
) -> Dict[str, Dict]:
"""Return {ref_id: {invocations_30d, unique_users_30d, trend_pct}}.
One query per source per page render — avoids N+1.
"""
rows = conn.execute("""
WITH last_30 AS (
SELECT ref_id,
SUM(invocations) AS inv30,
SUM(distinct_users) AS u30
FROM usage_plugin_daily
WHERE source = ? AND day >= CURRENT_DATE - INTERVAL 30 DAY
GROUP BY ref_id
),
prior_week AS (
SELECT ref_id, SUM(invocations) AS inv_prior
FROM usage_plugin_daily
WHERE source = ?
AND day >= CURRENT_DATE - INTERVAL 14 DAY
AND day < CURRENT_DATE - INTERVAL 7 DAY
GROUP BY ref_id
),
recent_week AS (
SELECT ref_id, SUM(invocations) AS inv_recent
FROM usage_plugin_daily
WHERE source = ?
AND day >= CURRENT_DATE - INTERVAL 7 DAY
GROUP BY ref_id
)
SELECT l.ref_id, l.inv30, l.u30, p.inv_prior, r.inv_recent
FROM last_30 l
LEFT JOIN prior_week p USING (ref_id)
LEFT JOIN recent_week r USING (ref_id)
""", [source, source, source]).fetchall()
out: Dict[str, Dict] = {}
for ref_id, inv30, u30, prior, recent in rows:
trend = None
if prior is not None and prior >= 3:
_recent = recent or 0
trend = (_recent - prior) / prior * 100.0
out[ref_id] = {
"invocations_30d": int(inv30 or 0),
"unique_users_30d": int(u30 or 0),
"trend_pct": trend,
}
return out
def _load_plugin_daily_series(
conn: duckdb.DuckDBPyConnection,
source: str,
ref_id: str,
) -> List[Dict]:
"""Return a 30-entry list [{day, invocations}] with missing days filled to 0."""
rows = conn.execute("""
SELECT day, SUM(invocations) AS inv
FROM usage_plugin_daily
WHERE source = ? AND ref_id = ?
AND day >= CURRENT_DATE - INTERVAL 30 DAY
GROUP BY day
ORDER BY day
""", [source, ref_id]).fetchall()
by_day = {str(r[0]): int(r[1] or 0) for r in rows}
import datetime as _dt
today = _dt.date.today()
series = []
for offset in range(29, -1, -1):
day = today - _dt.timedelta(days=offset)
day_str = day.isoformat()
series.append({"day": day_str, "invocations": by_day.get(day_str, 0)})
return series
def _audit(
conn: duckdb.DuckDBPyConnection,
actor_id: str,
action: str,
target: str,
params: Optional[dict] = None,
) -> None:
try:
AuditRepository(conn).log(
user_id=actor_id, action=action, resource=target, params=params
)
except Exception:
pass
def _invalidate_etag() -> None:
try:
from app.marketplace_server import packager
packager.invalidate_etag_cache()
except Exception:
logger.exception("failed to invalidate marketplace etag cache")
def _frontmatter_body(text: str) -> str:
m = _FRONTMATTER_RE.match(text)
return text[m.end():].lstrip("\n") if m else text
def _to_iso(value: Any) -> Optional[str]:
if value is None:
return None
try:
return value.isoformat()
except AttributeError:
return str(value)
def _icon_key_for(category: Optional[str]) -> str:
"""Stable key used by the frontend to look up the inline SVG icon
(registered in ``src/category_icons.py``). Matches our taxonomy 1:1."""
return category or "Other"
def _curated_detail_url(marketplace_id: str, plugin_name: str) -> str:
return f"/marketplace/curated/{marketplace_id}/{plugin_name}"
def _flea_detail_url(entity_id: str) -> str:
return f"/marketplace/flea/{entity_id}"
def _resolve_marketplace_meta(
conn: duckdb.DuckDBPyConnection, marketplace_id: str
) -> Dict[str, Optional[str]]:
"""Look up display name + curator metadata in one DB hit.
Returns a dict with keys ``name``, ``curator_name``, ``curator_email``.
Missing rows fall back to the marketplace_id as ``name`` and ``None``
for the curator fields. Caller decides what to do with the absence
(typically: surface the ``OWNER_TODO_PLACEHOLDER``).
"""
row = MarketplaceRegistryRepository(conn).get(marketplace_id)
if not row:
return {
"name": marketplace_id,
"curator_name": None,
"curator_email": None,
}
name = (row.get("name") or "").strip() or marketplace_id
return {
"name": name,
"curator_name": (row.get("curator_name") or "").strip() or None,
"curator_email": (row.get("curator_email") or "").strip() or None,
}
def _curated_to_item(
conn: duckdb.DuckDBPyConnection,
plugin_row: dict,
*,
subs: set,
marketplace_meta: Dict[str, Dict[str, Optional[str]]],
stats: Optional[Dict[str, Dict]] = None,
) -> MarketplaceItem:
marketplace_id = plugin_row["marketplace_id"]
plugin_name = plugin_row["name"]
meta = marketplace_meta.get(marketplace_id) or {
"name": marketplace_id,
"curator_name": None,
"curator_email": None,
}
# v32: card "owner" is the curator from the marketplace registry — not the
# upstream `marketplace.json::author.name` we historically used. Curator
# is the human accountable for the plugin's presence in this Agnes
# instance; upstream author belongs in the plugin source repo.
owner = meta["curator_name"] or OWNER_TODO_PLACEHOLDER
# Lazy enrichment read for the listing card — cached per (marketplace_id,
# mtime) so a page of 24 plugins from the same marketplace triggers ONE
# disk read, not 24. Empty dict when curator hasn't filled the fields →
# listing card falls back to raw name + marketplace.json description.
enrichment = _curated_plugin_enrichment(marketplace_id, plugin_name)
ref_id = f"{marketplace_id}/{plugin_name}"
stat = (stats or {}).get(ref_id, {})
return MarketplaceItem(
id=f"curated-{marketplace_id}/{plugin_name}",
source="curated",
name=plugin_name,
type="plugin",
category=plugin_row.get("category") or None,
description=plugin_row.get("description"),
owner=owner,
version=plugin_row.get("version"),
# Cover photo URL is already in served form (internal `/asset/`,
# mirrored `/mirrored/`, or pass-through external URL) — see
# src.marketplace._refresh_plugin_cache for the resolution path.
photo_url=plugin_row.get("cover_photo_url"),
added=_to_iso(plugin_row.get("created_at")),
installed=(marketplace_id, plugin_name) in subs,
marketplace_slug=marketplace_id,
marketplace_name=meta["name"],
detail_url=_curated_detail_url(marketplace_id, plugin_name),
is_system=bool(plugin_row.get("is_system")),
display_name=enrichment.get("display_name"),
tagline=enrichment.get("tagline"),
invocations_30d=stat.get("invocations_30d", 0),
unique_users_30d=stat.get("unique_users_30d", 0),
trend_pct=stat.get("trend_pct"),
)
def _flea_to_item(
entity: dict,
*,
installed_set: set,
viewer_id: Optional[str] = None,
stats: Optional[Dict[str, Dict]] = None,
) -> MarketplaceItem:
photo_url = (
f"/api/store/entities/{entity['id']}/photo"
if entity.get("photo_path") else None
)
# The archive flow renames the row's `name` to free the slot; strip
# the suffix when rendering listings so owners don't see the ugly
# `__archived__<epoch>` in their own cards. The served catalog
# (Claude Code's `/plugin` resolution) uses the renamed slug — we
# don't strip there.
from src.store_naming import strip_archive_suffix
display_name = strip_archive_suffix(entity["name"])
invocation = suffixed_name(display_name, entity.get("owner_username") or "")
is_viewer_owner = bool(viewer_id and entity.get("owner_user_id") == viewer_id)
stat = (stats or {}).get(entity["id"], {})
return MarketplaceItem(
id=f"flea-{entity['id']}",
source="flea",
name=invocation,
type=entity["type"],
category=entity.get("category") or None,
description=entity.get("description"),
owner=entity.get("owner_username"),
version=entity.get("version"),
photo_url=photo_url,
added=_to_iso(entity.get("created_at")),
installed=entity["id"] in installed_set,
marketplace_slug=None,
marketplace_name=None,
detail_url=_flea_detail_url(entity["id"]),
visibility_status=entity.get("visibility_status") or "approved",
is_viewer_owner=is_viewer_owner,
invocations_30d=stat.get("invocations_30d", 0),
unique_users_30d=stat.get("unique_users_30d", 0),
trend_pct=stat.get("trend_pct"),
)
# ---------------------------------------------------------------------------
# GET /api/marketplace/items
# ---------------------------------------------------------------------------
def _build_telemetry(
conn: duckdb.DuckDBPyConnection,
source: str,
ref_id: str,
) -> Optional[Dict[str, Any]]:
"""Build the telemetry dict for detail endpoints.
Returns None when invocations_30d == 0 (no data yet).
Otherwise returns {invocations_30d, unique_users_30d, daily_series}.
"""
stats = _load_invocation_stats(conn, source)
stat = stats.get(ref_id)
inv30 = stat["invocations_30d"] if stat else 0
if inv30 == 0:
return None
return {
"invocations_30d": inv30,
"unique_users_30d": stat["unique_users_30d"] if stat else 0,
"daily_series": _load_plugin_daily_series(conn, source, ref_id),
}
def _apply_sort(
items: List[MarketplaceItem],
sort: str,
) -> List[MarketplaceItem]:
"""Sort a list of MarketplaceItem objects in-place and return it.
- ``recent`` — preserve existing order (no-op).
- ``most_used`` — DESC by invocations_30d, then DESC install_count, then name ASC.
- ``trending`` — DESC by trend_pct; items with trend_pct=None are excluded.
"""
if sort == "most_used":
items.sort(
key=lambda it: (-it.invocations_30d, it.name.lower())
)
elif sort == "trending":
items = [it for it in items if it.trend_pct is not None]
items.sort(key=lambda it: -(it.trend_pct or 0.0))
return items
@router.get("/items", response_model=ItemListResponse)
async def list_items(
tab: Literal["curated", "flea", "my"] = Query("curated"),
q: Optional[str] = Query(None),
category: Optional[str] = Query(None),
type: Optional[Literal["skill", "agent", "plugin"]] = Query(None),
sort: Literal["recent", "most_used", "trending"] = Query("recent"),
page: int = Query(1, ge=1),
page_size: int = Query(24, ge=1, le=100),
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
"""Paginated, RBAC-scoped item list for the ``/marketplace`` browse page.
Per-tab dispatch:
* ``curated`` → ``MarketplacePluginsRepository.list_with_filters`` scoped to
the caller's ``user_group_members``. Results are tagged with the
caller's subscription state.
* ``flea`` → ``StoreEntitiesRepository.list`` (community Store).
* ``my`` → direct read from ``user_curated_subscriptions`` (filtered
against the caller's RBAC-allowed plugins) ``user_store_installs``
(all flea types — skill / agent / plugin — surface as individual
cards). We deliberately don't reuse ``resolve_user_marketplace``
here: that resolver bundles flea skills/agents into a single
synthetic ``store-bundle`` entry useful for the served Claude Code
marketplace ZIP/git endpoints but wrong for /marketplace browsing,
where each item should appear as its own card. Mirrors the
``/api/my-stack`` reading pattern.
``sort`` controls ordering after stats are joined:
* ``recent`` — existing DB order (default, backward-compatible).
* ``most_used`` — DESC by invocations_30d, ties by install_count then name.
* ``trending`` — DESC by trend_pct; items with no trend data are excluded.
"""
skip = (page - 1) * page_size
needs_sort = sort != "recent"
if tab == "curated":
group_ids = _user_group_ids(user["id"], conn) or set()
subs = UserCuratedSubscriptionsRepository(conn).subscribed_set(user["id"])
# When sorting, we need all rows to sort before paginating.
if needs_sort:
all_rows, total = MarketplacePluginsRepository(conn).list_with_filters(
group_ids=group_ids,
search=q or None,
category=category or None,
skip=0,
limit=10000,
)
else:
all_rows, total = MarketplacePluginsRepository(conn).list_with_filters(
group_ids=group_ids,
search=q or None,
category=category or None,
skip=skip,
limit=page_size,
)
marketplace_meta: Dict[str, Dict[str, Optional[str]]] = {}
if all_rows:
distinct_ids = {r["marketplace_id"] for r in all_rows}
for mp_id in distinct_ids:
marketplace_meta[mp_id] = _resolve_marketplace_meta(conn, mp_id)
curated_stats = _load_invocation_stats(conn, "curated")
items = [
_curated_to_item(
conn, r, subs=subs,
marketplace_meta=marketplace_meta,
stats=curated_stats,
)
for r in all_rows
]
if needs_sort:
items = _apply_sort(items, sort)
total = len(items)
items = items[skip: skip + page_size]
return ItemListResponse(
items=items, total=total, page=page, page_size=page_size,
)
if tab == "flea":
installed_set = {
row["id"]
for row in UserStoreInstallsRepository(conn).list_for_user(user["id"])
}
# Visibility filter: non-admin sees approved + their own
# non-approved (so submitters spot what's still under review
# in their own grid). Admin sees everything.
from app.auth.access import is_user_admin
if is_user_admin(user["id"], conn):
visibility_filter = None
include_owner = None
else:
visibility_filter = ["approved"]
include_owner = user["id"]
if needs_sort:
all_flea_rows, total = StoreEntitiesRepository(conn).list(
skip=0, limit=10000,
type=type, category=category, search=q or None,
visibility_status=visibility_filter,
include_owner_id=include_owner,
)
else:
all_flea_rows, total = StoreEntitiesRepository(conn).list(
skip=skip, limit=page_size,
type=type, category=category, search=q or None,
visibility_status=visibility_filter,
include_owner_id=include_owner,
)
flea_stats = _load_invocation_stats(conn, "flea")
items = [
_flea_to_item(
r, installed_set=installed_set,
viewer_id=user["id"],
stats=flea_stats,
)
for r in all_flea_rows
]
if needs_sort:
items = _apply_sort(items, sort)
total = len(items)
items = items[skip: skip + page_size]
return ItemListResponse(
items=items, total=total, page=page, page_size=page_size,
)
# tab == "my" — see docstring; read directly from source-of-truth tables.
items: List[MarketplaceItem] = []
granted = resolve_allowed_plugins(conn, user)
subs = UserCuratedSubscriptionsRepository(conn).subscribed_set(user["id"])
marketplace_meta: Dict[str, Dict[str, Optional[str]]] = {}
# Pull the enriched rows the Curated tab uses (cover_photo_url, video_url,
# category override, doc_links from marketplace-metadata.json) so the My Stack
# cards look identical to the curated cards the user just clicked
# "+ Add to my stack" on. ``resolve_allowed_plugins`` reads only the
# upstream marketplace.json, which doesn't carry those columns; without
# this lookup the same plugin renders with its cover photo on
# ``?tab=curated`` and with a gradient placeholder on ``?tab=my``.
plugin_repo = MarketplacePluginsRepository(conn)
subscribed_mp_ids = {mp_id for (mp_id, _) in subs}
enriched_lookup: Dict[Tuple[str, str], Dict[str, Any]] = {}
for mp_id in subscribed_mp_ids:
for row in plugin_repo.list_for_marketplace(mp_id):
enriched_lookup[(mp_id, row["name"])] = row
curated_stats = _load_invocation_stats(conn, "curated")
flea_stats = _load_invocation_stats(conn, "flea")
for p in granted:
key = (p["marketplace_id"], p["original_name"])
if key not in subs:
continue
mp_id = p["marketplace_id"]
if mp_id not in marketplace_meta:
marketplace_meta[mp_id] = _resolve_marketplace_meta(conn, mp_id)
plugin_row = enriched_lookup.get(key)
if plugin_row is None:
# Fallback: plugin in RBAC + subscribed but not yet ingested into
# marketplace_plugins (rare race — granted before the first sync
# cycle runs). Build the bare shape from the on-disk manifest so
# the card still renders, just without marketplace-metadata enrichment;
# cover falls through to the gradient placeholder until the next
# sync.
author = p["raw"].get("author")
plugin_row = {
"marketplace_id": mp_id,
"name": p["original_name"],
"description": p["raw"].get("description"),
"version": p.get("version"),
"category": p["raw"].get("category"),
"author_name": author.get("name") if isinstance(author, dict) else None,
"cover_photo_url": None,
"created_at": None,
}
items.append(_curated_to_item(
conn, plugin_row, subs=subs, marketplace_meta=marketplace_meta,
stats=curated_stats,
))
flea_installs = UserStoreInstallsRepository(conn).list_for_user(user["id"])
flea_installed_set = {row["id"] for row in flea_installs}
for entity in flea_installs:
items.append(_flea_to_item(
entity, installed_set=flea_installed_set, stats=flea_stats,
))
# Apply optional filters client-server-style for `my` tab (small N):
if q:
needle = q.lower()
items = [
it for it in items
if needle in it.name.lower()
or needle in (it.description or "").lower()
or needle in (it.owner or "").lower()
or needle in (it.category or "").lower()
]
if category:
items = [it for it in items if (it.category or "Other") == category]
if type:
items = [it for it in items if it.type == type]
items = _apply_sort(items, sort)
total = len(items)
items = items[skip : skip + page_size]
return ItemListResponse(
items=items, total=total, page=page, page_size=page_size,
)
# ---------------------------------------------------------------------------
# GET /api/marketplace/categories
# ---------------------------------------------------------------------------
@router.get("/categories", response_model=CategoriesResponse)
async def list_categories(
tab: Literal["curated", "flea", "my"] = Query("curated"),
type: Optional[Literal["skill", "agent", "plugin"]] = Query(None),
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
"""Per-tab category list with non-zero counts.
Source of categories: union of ``STORE_CATEGORIES`` and any non-empty
``marketplace_plugins.category`` values in the caller's RBAC scope.
Categories with zero matching items are omitted (the frontend hides
them this way).
"""
counts: Dict[str, int] = {}
if tab in ("curated", "my"):
group_ids = _user_group_ids(user["id"], conn) or set()
if tab == "curated":
counts.update(
MarketplacePluginsRepository(conn).category_counts(
group_ids=group_ids
)
)
else: # my — direct read mirroring the items endpoint's `my` branch.
granted = resolve_allowed_plugins(conn, user)
subs = UserCuratedSubscriptionsRepository(conn).subscribed_set(user["id"])
# Curated plugins are always type='plugin'. When the type filter
# is set to skill/agent, those rows won't show in the items grid
# (filtered out by the items endpoint at line 579), so they must
# not contribute to the category counts either — otherwise the
# pill counts overstate what the user will actually see.
if type is None or type == "plugin":
for p in granted:
if (p["marketplace_id"], p["original_name"]) not in subs:
continue
cat = (p["raw"].get("category") or "").strip() or "Other"
counts[cat] = counts.get(cat, 0) + 1
for row in UserStoreInstallsRepository(conn).list_for_user(user["id"]):
if type and row.get("type") != type:
continue
cat = (row.get("category") or "").strip() or "Other"
counts[cat] = counts.get(cat, 0) + 1
if tab == "flea":
# Visibility filter (v32+/v35): non-admin counts approved + own
# non-archived non-approved (mirrors the listing endpoint so the
# category counts match what the user actually sees in the
# grid). Admin counts everything.
from app.auth.access import is_user_admin
clauses: List[str] = []
sql_params: List[Any] = []
if type:
clauses.append("type = ?"); sql_params.append(type)
if not is_user_admin(user["id"], conn):
clauses.append(
"(visibility_status = 'approved' "
"OR (owner_user_id = ? AND visibility_status != 'archived'))"
)
sql_params.append(user["id"])
where_sql = ("WHERE " + " AND ".join(clauses)) if clauses else ""
rows = conn.execute(
f"SELECT COALESCE(NULLIF(TRIM(category),''), 'Other') AS cat, COUNT(*) "
f"FROM store_entities {where_sql} GROUP BY cat",
sql_params,
).fetchall()
for r in rows:
counts[str(r[0])] = counts.get(str(r[0]), 0) + int(r[1])
items = [
CategoryEntry(name=name, count=count, icon_key=_icon_key_for(name))
for name, count in sorted(counts.items())
if count > 0
]
return CategoriesResponse(items=items)
# ---------------------------------------------------------------------------
# Curated detail + install endpoints
# ---------------------------------------------------------------------------
def _list_inner_skills(plugin_root: Path) -> List[InnerItemSummary]:
"""Build ``InnerItemSummary`` objects from the shared listing helper.
The shared helper (``src.marketplace_listing.list_inner_skills``) returns
plain names; the API layer re-reads each SKILL.md to populate the
``description`` field used on the detail page.
"""
out: List[InnerItemSummary] = []
skills_dir = plugin_root / "skills"
if not skills_dir.is_dir():
return out
for skill_dir in sorted(skills_dir.iterdir()):
if not skill_dir.is_dir():
continue
skill_md = skill_dir / "SKILL.md"
if not skill_md.is_file():
continue
try:
text = skill_md.read_text(encoding="utf-8", errors="replace")
except OSError:
continue
fm = _parse_frontmatter(text)
out.append(InnerItemSummary(
name=fm.get("name") or skill_dir.name,
description=fm.get("description"),
detail_url=None, # populated by caller (needs marketplace_id + plugin_name)
))
return out
def _list_inner_agents(plugin_root: Path) -> List[InnerItemSummary]:
"""Build ``InnerItemSummary`` objects from the shared listing helper."""
out: List[InnerItemSummary] = []
agents_dir = plugin_root / "agents"
if not agents_dir.is_dir():
return out
for agent_path in sorted(agents_dir.iterdir()):
if not agent_path.is_file() or agent_path.suffix != ".md":
continue
try:
text = agent_path.read_text(encoding="utf-8", errors="replace")
except OSError:
continue
fm = _parse_frontmatter(text)
out.append(InnerItemSummary(
name=fm.get("name") or agent_path.stem,
description=fm.get("description"),
detail_url=None,
))
return out
def _list_commands(plugin_root: Path) -> List[CommandEntry]:
"""Return ``commands/*.md`` as ``CommandEntry`` objects from frontmatter.
Names from the shared ``src.marketplace_listing.list_commands`` helper
already carry the leading ``/``; re-read here to pick up ``description``.
"""
d = plugin_root / "commands"
if not d.is_dir():
return []
out: List[CommandEntry] = []
for p in sorted(d.iterdir()):
if not p.is_file() or p.suffix != ".md":
continue
try:
text = p.read_text(encoding="utf-8", errors="replace")
except OSError:
continue
fm = _parse_frontmatter(text)
raw = (fm.get("name") or p.stem or "").strip()
if not raw:
continue
name = raw if raw.startswith("/") else f"/{raw}"
out.append(CommandEntry(
name=name,
description=fm.get("description"),
))
return out
def _read_plugin_json(plugin_root: Path) -> dict:
"""Best-effort load of the plugin's own .claude-plugin/plugin.json."""
pj = plugin_root / ".claude-plugin" / "plugin.json"
if not pj.is_file():
return {}
try:
import json
data = json.loads(pj.read_text(encoding="utf-8"))
except (OSError, ValueError):
return {}
return data if isinstance(data, dict) else {}
def _list_hooks(plugin_root: Path) -> List[HookEntry]:
"""Combine hook declarations from plugin.json with files in ``hooks/``.
Two declaration paths exist:
1. ``plugin.json`` ``hooks`` field — ``{<EventName>: [{matcher, hooks: [...]}]}``
2. Bare scripts in ``<plugin_root>/hooks/`` — event unknown, rendered
as ``""`` so the table still surfaces them.
"""
out: List[HookEntry] = []
seen_names: set[str] = set()
# Path 1: structured declaration in plugin.json.
data = _read_plugin_json(plugin_root)
hooks_field = data.get("hooks") if isinstance(data, dict) else None
if isinstance(hooks_field, dict):
for event, entries in hooks_field.items():
if not isinstance(entries, list):
continue
for entry in entries:
if not isinstance(entry, dict):
continue
inner = entry.get("hooks") or []
if not isinstance(inner, list):
continue
for h in inner:
if not isinstance(h, dict):
continue
name = h.get("command") or h.get("name") or h.get("type") or "(hook)"
name = str(name).strip() or "(hook)"
description = entry.get("matcher") or h.get("description")
out.append(HookEntry(
name=name,
event=str(event),
description=str(description) if description else None,
))
seen_names.add(name)
# Path 2: bare scripts in hooks/ directory not already covered.
hooks_dir = plugin_root / "hooks"
if hooks_dir.is_dir():
for p in sorted(hooks_dir.iterdir()):
if not p.is_file():
continue
if p.name in seen_names or str(p) in seen_names:
continue
out.append(HookEntry(name=p.name, event=None, description=None))
return out
def _list_mcps(plugin_root: Path) -> List[McpEntry]:
"""Parse MCP server declarations from ``.mcp.json`` (or legacy
``mcp_servers.json``). Returns ``[(name, type, description)]``.
Standard shape: ``{"mcpServers": {<name>: {<config>}}}`` where ``<config>``
contains either ``command`` (stdio) or ``url`` (sse). Type is inferred
from those fields. Description falls through to ``command`` so the
operator at least sees what's launched.
"""
candidates = [
plugin_root / ".mcp.json",
plugin_root / "mcp_servers.json",
]
out: List[McpEntry] = []
for path in candidates:
if not path.is_file():
continue
try:
import json
data = json.loads(path.read_text(encoding="utf-8"))
except (OSError, ValueError):
continue
servers = data.get("mcpServers") if isinstance(data, dict) else None
if not isinstance(servers, dict):
continue
for name, cfg in servers.items():
if not isinstance(cfg, dict):
out.append(McpEntry(name=str(name)))
continue
kind: Optional[str] = None
description: Optional[str] = None
if "command" in cfg:
kind = "stdio"
description = str(cfg.get("command"))
args = cfg.get("args")
if isinstance(args, list) and args:
description = description + " " + " ".join(str(a) for a in args)
elif "url" in cfg:
kind = "sse"
description = str(cfg.get("url"))
out.append(McpEntry(
name=str(name),
type=kind,
description=description,
))
break # first found candidate wins
return out
def _bundle_size(plugin_root: Path) -> Optional[int]:
"""Sum of file sizes under ``plugin_root``. None if path missing."""
if not plugin_root.is_dir():
return None
total = 0
for p in plugin_root.rglob("*"):
if p.is_file():
try:
total += p.stat().st_size
except OSError:
pass
return total
def _walk_files(root: Path) -> List[FileEntry]:
"""Recursive file listing under ``root``.
Returns sorted ``[FileEntry(path, size)]`` with paths relative to ``root``
(forward-slash form for stable display). Directories are skipped — only
leaf files appear, matching the way ``store_detail.html`` renders the
Files section. Empty list if ``root`` is missing or not a directory.
"""
if not root.is_dir():
return []
out: List[FileEntry] = []
for p in sorted(root.rglob("*")):
if not p.is_file():
continue
try:
rel = p.relative_to(root).as_posix()
out.append(FileEntry(path=rel, size=p.stat().st_size))
except OSError:
continue
return out
def _resolve_owner_display(
conn: duckdb.DuckDBPyConnection,
owner_user_id: str,
fallback: str,
) -> str:
"""Friendly owner display name — ``users.name → users.email → fallback``.
Mirrors the inline lookup ``app/web/router.py::store_detail`` already does
so the marketplace API surfaces the same string the Store page shows.
"""
row = conn.execute(
"SELECT name, email FROM users WHERE id = ?", [owner_user_id]
).fetchone()
if not row:
return fallback
return row[0] or row[1] or fallback
def _get_plugin_row(
conn: duckdb.DuckDBPyConnection,
marketplace_id: str,
plugin_name: str,
) -> Optional[dict]:
"""Fetch a single ``marketplace_plugins`` row as a dict, or ``None``."""
rows = conn.execute(
"SELECT * FROM marketplace_plugins "
"WHERE marketplace_id = ? AND name = ?",
[marketplace_id, plugin_name],
).fetchall()
if not rows:
return None
columns = [d[0] for d in conn.description]
return MarketplacePluginsRepository._row_to_dict(columns, rows[0])
@router.get(
"/curated/{marketplace_id}/{plugin_name}",
response_model=PluginDetailResponse,
)
async def curated_detail(
marketplace_id: str,
plugin_name: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
"""Return the curated plugin detail + inner skill/agent/command/hook/mcp list.
The 403 guard fires before this body runs (via ``require_resource_access``).
A second ``get_current_user`` dependency is included so we still have the
caller's user dict for the ``installed`` flag.
"""
plugin_row = _get_plugin_row(conn, marketplace_id, plugin_name)
if plugin_row is None:
raise HTTPException(status_code=404, detail="plugin_not_found")
plugin_root = (
Path(get_marketplaces_dir()) / marketplace_id / "plugins" / plugin_name
)
skills = _list_inner_skills(plugin_root)
agents = _list_inner_agents(plugin_root)
# v32: enrich each skill/agent card with its marketplace-metadata cover photo
# so the inner cards on the plugin detail page render the real image
# instead of just initials. Read marketplace-metadata + mirror manifest once
# (mtime cache makes the metadata read free on cache hit) and reuse for
# all inner items in the plugin.
inner_metadata = _read_metadata_cached(marketplace_id)
inner_manifest = _load_mirror_manifest(marketplace_id)
for s in skills:
s.detail_url = (
f"/marketplace/curated/{marketplace_id}/{plugin_name}/skill/{s.name}"
)
s.cover_photo_url = _curated_inner_cover(
marketplace_id, plugin_name, "skill", s.name,
manifest=inner_manifest, metadata=inner_metadata,
)
for a in agents:
a.detail_url = (
f"/marketplace/curated/{marketplace_id}/{plugin_name}/agent/{a.name}"
)
a.cover_photo_url = _curated_inner_cover(
marketplace_id, plugin_name, "agent", a.name,
manifest=inner_manifest, metadata=inner_metadata,
)
commands = _list_commands(plugin_root)
hooks = _list_hooks(plugin_root)
mcps = _list_mcps(plugin_root)
subs = UserCuratedSubscriptionsRepository(conn).subscribed_set(user["id"])
raw = plugin_row.get("raw") or {}
if isinstance(raw, str):
try:
import json
raw = json.loads(raw)
except (ValueError, TypeError):
raw = {}
# v32: cover, video, and doc_links come from `marketplace-metadata.json` via the
# sync pipeline (`src/marketplace.py::_refresh_plugin_cache`) — already in
# served-URL form when written to the DB. We pass them through the
# response model unchanged. Curator name + email come from the marketplace
# registry and surface as the plugin's accountable owner; the upstream
# `marketplace.json::author.name` is intentionally not used here so we
# have a single source of truth for "who runs this".
meta = _resolve_marketplace_meta(conn, marketplace_id)
doc_link_entries: List[DocEntry] = []
raw_links = plugin_row.get("doc_links")
if isinstance(raw_links, list):
for link in raw_links:
if isinstance(link, dict) and link.get("name") and link.get("url"):
doc_link_entries.append(
DocEntry(name=str(link["name"]), url=str(link["url"]))
)
# Plugin-level rich content (display_name, tagline, description_long_html,
# use_cases, sample_interaction) — on-demand read from the working tree's
# marketplace-metadata.json so curator edits land at next page refresh
# without a sync cycle. Empty dict when curator hasn't filled any of the
# new fields → PluginDetailResponse renders the historical fallback shape.
enrichment = _curated_plugin_enrichment(marketplace_id, plugin_name)
return PluginDetailResponse(
source="curated",
marketplace_id=marketplace_id,
marketplace_name=meta["name"],
plugin_name=plugin_name,
manifest_name=(raw.get("name") if isinstance(raw, dict) else None) or plugin_name,
description=plugin_row.get("description"),
version=plugin_row.get("version"),
category=plugin_row.get("category"),
author_name=meta["curator_name"] or OWNER_TODO_PLACEHOLDER,
curator_email=meta["curator_email"],
homepage=plugin_row.get("homepage"),
cover_photo_url=plugin_row.get("cover_photo_url"),
video_url=plugin_row.get("video_url"),
bundle_size=_bundle_size(plugin_root),
released_at=_to_iso(plugin_row.get("created_at")),
updated_at=_to_iso(plugin_row.get("updated_at")),
installed=(marketplace_id, plugin_name) in subs,
skills=skills,
agents=agents,
commands=commands,
hooks=hooks,
mcps=mcps,
files=_walk_files(plugin_root),
docs=doc_link_entries,
is_system=bool(plugin_row.get("is_system")),
**enrichment,
telemetry=_build_telemetry(
conn, "curated", f"{marketplace_id}/{plugin_name}",
),
)
@router.get(
"/flea/{entity_id}/detail",
response_model=PluginDetailResponse,
)
async def flea_detail(
entity_id: str,
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
"""Same shape as curated detail, sourced from a Store entity.
Flea entities live at ``${DATA_DIR}/store/<entity_id>/plugin/`` —
canonical Claude Code plugin tree, so the same parsers apply.
"""
from app.api.store import _enforce_visibility
from app.utils import get_store_dir
entity = StoreEntitiesRepository(conn).get(entity_id)
if not entity:
raise HTTPException(status_code=404, detail="entity_not_found")
# Same gate as /api/store/entities/{id}: non-owner non-admin gets
# 404 (not 403) for any non-approved entity. Without this, a user
# who guesses an entity_id can still pull the bundle metadata
# through the marketplace JSON feed even though it's quarantined
# and excluded from the public listing.
_enforce_visibility(entity, user, conn)
plugin_root = Path(get_store_dir()) / entity_id / "plugin"
if entity["type"] == "plugin":
skills = _list_inner_skills(plugin_root)
for s in skills:
s.detail_url = f"/marketplace/flea/{entity_id}/skill/{s.name}"
agents = _list_inner_agents(plugin_root)
for a in agents:
a.detail_url = f"/marketplace/flea/{entity_id}/agent/{a.name}"
commands = _list_commands(plugin_root)
hooks = _list_hooks(plugin_root)
mcps = _list_mcps(plugin_root)
else:
# skill / agent — inner structure isn't applicable; the dedicated
# item-detail page will render those entities. We still return the
# response so the frontend can decide which template to use.
skills = []
agents = []
commands = []
hooks = []
mcps = []
is_installed = UserStoreInstallsRepository(conn).is_installed(
user["id"], entity_id,
)
cover_url: Optional[str] = None
if entity.get("photo_path"):
cover_url = f"/api/store/entities/{entity_id}/photo"
# Strip archive-rename suffix for human display; manifest_name keeps
# the renamed-on-archive slug since that's what Claude Code resolves.
from src.store_naming import strip_archive_suffix
_flea_display_name = strip_archive_suffix(entity["name"])
invocation = suffixed_name(_flea_display_name, entity.get("owner_username") or "")
# doc_paths is a JSON array of relative paths the uploader picked at upload
# time; `app/api/store.py` serves them by basename via /api/store/.../docs/{filename}.
docs: List[DocEntry] = []
for raw_path in (entity.get("doc_paths") or []):
if not isinstance(raw_path, str) or not raw_path.strip():
continue
fname = Path(raw_path).name
if not fname:
continue
docs.append(DocEntry(
name=fname,
url=f"/api/store/entities/{entity_id}/docs/{fname}",
))
owner_display = _resolve_owner_display(
conn,
entity["owner_user_id"],
entity.get("owner_username") or "",
)
# Surface the latest submission verdict to the owner / admin so the
# quarantine banner's auto-refresh JS has a signal to poll on.
# Visibility alone is not enough because `blocked_llm` keeps the
# entity at `visibility_status='pending'`.
submission_status: Optional[str] = None
is_owner = entity.get("owner_user_id") == user.get("id")
is_admin_user = is_user_admin(user["id"], conn)
if is_owner or is_admin_user:
from src.repositories.store_submissions import StoreSubmissionsRepository
latest_sub = StoreSubmissionsRepository(conn).latest_for_entity(entity_id)
if latest_sub:
submission_status = latest_sub.get("status")
return PluginDetailResponse(
source="flea",
entity_id=entity_id,
plugin_name=_flea_display_name,
manifest_name=invocation,
description=entity.get("description"),
version=entity.get("version"),
category=entity.get("category"),
author_name=entity.get("owner_username"),
owner_display=owner_display,
homepage=None,
cover_photo_url=cover_url,
bundle_size=int(entity.get("file_size") or 0) or _bundle_size(plugin_root),
install_count=int(entity.get("install_count") or 0),
released_at=_to_iso(entity.get("created_at")),
updated_at=_to_iso(entity.get("updated_at")),
installed=is_installed,
skills=skills,
agents=agents,
commands=commands,
hooks=hooks,
mcps=mcps,
files=_walk_files(plugin_root),
docs=docs,
visibility_status=entity.get("visibility_status") or "approved",
submission_status=submission_status,
telemetry=_build_telemetry(conn, "flea", entity_id),
)
@router.post(
"/curated/{marketplace_id}/{plugin_name}/install",
response_model=InstallActionResponse,
)
async def curated_install(
marketplace_id: str,
plugin_name: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
"""Subscribe the caller to a curated plugin (Model B opt-in).
Idempotent — repeated calls are no-ops. The plugin must exist in
``marketplace_plugins``; otherwise 404. The RBAC guard already ensured
the caller is allowed to install.
"""
exists = conn.execute(
"SELECT 1 FROM marketplace_plugins WHERE marketplace_id = ? AND name = ?",
[marketplace_id, plugin_name],
).fetchone()
if not exists:
raise HTTPException(status_code=404, detail="plugin_not_found")
inserted = UserCuratedSubscriptionsRepository(conn).subscribe(
user["id"], marketplace_id, plugin_name,
)
if inserted:
_audit(
conn, user["id"], "marketplace.curated.install",
f"plugin:{marketplace_id}/{plugin_name}",
)
_invalidate_etag()
return InstallActionResponse(installed=True)
@router.delete(
"/curated/{marketplace_id}/{plugin_name}/install",
response_model=InstallActionResponse,
)
async def curated_uninstall(
marketplace_id: str,
plugin_name: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
user: dict = Depends(get_current_user),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
# v39: system plugins are mandatory for every user — refuse uninstall.
sys_row = conn.execute(
"SELECT is_system FROM marketplace_plugins "
"WHERE marketplace_id = ? AND name = ?",
[marketplace_id, plugin_name],
).fetchone()
if sys_row and bool(sys_row[0]):
raise HTTPException(
status_code=409,
detail="cannot_uninstall_system_plugin",
)
deleted = UserCuratedSubscriptionsRepository(conn).unsubscribe(
user["id"], marketplace_id, plugin_name,
)
if deleted:
_audit(
conn, user["id"], "marketplace.curated.uninstall",
f"plugin:{marketplace_id}/{plugin_name}",
)
_invalidate_etag()
return InstallActionResponse(installed=False)
# ---------------------------------------------------------------------------
# Curated inner skill/agent detail
# ---------------------------------------------------------------------------
def _safe_join(plugin_root: Path, *parts: str) -> Optional[Path]:
"""Join ``parts`` onto ``plugin_root`` and return the resolved path only if
it actually lives under ``plugin_root``. Defends against ``..`` segments,
Windows ``\\`` separators that slip past Starlette's ``[^/]+`` path-param
regex, and symlinks planted inside a curated marketplace's git mirror that
point outside the plugin's own tree.
Returns ``None`` when the candidate doesn't exist, can't be resolved, or
escapes ``plugin_root``. Callers translate that into a 404.
"""
candidate = plugin_root.joinpath(*parts)
try:
resolved = candidate.resolve(strict=True)
root = plugin_root.resolve(strict=True)
resolved.relative_to(root)
except (OSError, ValueError):
return None
return resolved
def _read_inner(
plugin_root: Path,
sub: str,
name: str,
*,
is_dir_layout: bool,
) -> Optional[tuple[str, str]]:
"""Return (text, relpath) for a skill (sub='skills', is_dir=True) or
agent (sub='agents', is_dir=False) inside the plugin tree, or None if
the file is missing or escapes ``plugin_root`` (see ``_safe_join``).
"""
if is_dir_layout:
candidate = _safe_join(plugin_root, sub, name, "SKILL.md")
else:
candidate = _safe_join(plugin_root, sub, f"{name}.md")
if candidate is None:
return None
try:
text = candidate.read_text(encoding="utf-8", errors="replace")
except OSError:
return None
try:
rel = candidate.relative_to(plugin_root.resolve()).as_posix()
except ValueError:
rel = candidate.name
return text, rel
def _curated_inner_parent_fields(
conn: duckdb.DuckDBPyConnection, marketplace_id: str, plugin_name: str,
) -> Dict[str, Any]:
"""Pull the parent-plugin metadata that the redesigned skill/agent detail
page renders in the hero badges, meta-row, and sidebar.
Returns a dict with ``marketplace_name``, ``category``, ``parent_author_name``,
``parent_updated_at``. Missing fields fall through to safe defaults so the
inner endpoint still succeeds when the parent row is absent (which would be
a sync-skew bug, but shouldn't 500 the page).
"""
plugin_row = _get_plugin_row(conn, marketplace_id, plugin_name) or {}
plugin_root = (
Path(get_marketplaces_dir()) / marketplace_id / "plugins" / plugin_name
)
meta = _resolve_marketplace_meta(conn, marketplace_id)
return {
"marketplace_name": meta["name"],
"category": plugin_row.get("category"),
"parent_author_name": meta["curator_name"] or OWNER_TODO_PLACEHOLDER,
"parent_updated_at": _to_iso(plugin_row.get("updated_at")),
"manifest_name": resolve_manifest_name(plugin_root, fallback=plugin_name),
}
def _load_mirror_manifest(marketplace_id: str) -> Dict[Tuple[str, str], Any]:
"""Read the asset-mirror manifest for one marketplace, keyed by
``(plugin_name, url)``.
Returns an empty dict when the cache directory or manifest doesn't exist
(fresh install, marketplace never synced) — callers treat that as "no
URL is mirrored" which collapses to the same path as a fetch failure.
"""
from src.marketplace_asset_mirror import _load_manifest
cache_dir = get_marketplace_cache_dir() / marketplace_id
return _load_manifest(cache_dir)
def _resolve_external_via_mirror(
marketplace_id: str,
plugin_name: str,
url: str,
manifest: Dict[Tuple[str, str], Any],
) -> Optional[str]:
"""Translate one external URL into the Agnes-served `/mirrored/` URL when
the asset mirror successfully cached it for THIS plugin; ``None`` otherwise.
Used by the inner-detail (skill/agent) enrichment to apply the same
"drop entries Agnes can't deliver" rule that the plugin-level sync
flow already enforces. When this returns None the caller drops the
enrichment entry entirely (no broken external link surfaces in the UI).
"""
entry = manifest.get((plugin_name, url))
if entry is None or entry.status != "ok" or not entry.local:
return None
# Manifest stores `local` as `<plugin>/<rest>`; the /mirrored/ endpoint
# expects just `<rest>` (the plugin segment is in the URL path). Same
# transform as src/marketplace.py uses on the plugin-level path.
rest = entry.local.split("/", 1)[1] if "/" in entry.local else entry.local
return mirrored_url(marketplace_id, plugin_name, rest)
def _safe_use_case(raw: Any) -> Optional[UseCase]:
"""Build a ``UseCase`` from curator JSON, skipping malformed entries.
Curator-authored input — a missing or empty ``title`` / ``description`` /
``prompt`` returns ``None`` so the caller drops the card instead of
500-ing the whole detail page on Pydantic's required-field validation.
"""
if not isinstance(raw, dict):
return None
title = raw.get("title")
description = raw.get("description")
prompt = raw.get("prompt")
if not (title and description and prompt):
return None
return UseCase(title=title, description=description, prompt=prompt)
def _safe_sample_interaction(raw: Any) -> Optional[SampleInteraction]:
"""Build a ``SampleInteraction`` from curator JSON, skipping malformed input.
Same rationale as :func:`_safe_use_case` — partial curator JSON should
silently drop the section, not crash the endpoint.
"""
from app.markdown_render import render_safe
if not isinstance(raw, dict):
return None
user = raw.get("user")
assistant = raw.get("assistant")
if not (user and assistant):
return None
return SampleInteraction(
user=user,
assistant=assistant,
assistant_html=render_safe(assistant),
)
def _curated_inner_enrichment(
marketplace_id: str,
plugin_name: str,
kind: str,
inner_name: str,
) -> Dict[str, Any]:
"""Load marketplace-metadata.json sub-tree for a single skill/agent.
Lives here (not in the sync pipeline) so curators can update marketplace-metadata
on a working tree and see the change at the next page refresh, without
waiting for a full plugin-cache rewrite. Read on every inner-detail
request — marketplace-metadata.json is small enough that disk hit cost is
negligible compared to the SKILL.md / agent.md frontmatter parse the
endpoint already does.
External URL handling matches the plugin-level sync flow:
* cover_photo external → kept only when the asset mirror has a
successful cached copy (URL resolves to /mirrored/...). Else dropped
→ the card / hero falls through to the gradient placeholder.
* doc_links external → same; entries that aren't mirrored OK get
dropped from the served list. Internal docs survive only when the
file actually exists in the working tree.
Returns a dict shaped for direct merge into the InnerDetailResponse:
``cover_photo_url`` (resolved served URL or None),
``video_url`` (str or None), ``docs`` (list of DocEntry).
"""
from src.marketplace_metadata import resolve_inner_metadata
repo_root = Path(get_marketplaces_dir()) / marketplace_id
# Route through the mtime cache so skill / agent detail hits don't
# re-parse the JSON on every request. Plugin listing already shares
# this cache, so opening 5 inner-detail pages on a marketplace whose
# metadata file hasn't changed = 0 disk reads beyond the first.
metadata = _read_metadata_cached(marketplace_id)
section_kind = "skills" if kind == "skill" else "agents"
resolved = resolve_inner_metadata(metadata, plugin_name, section_kind, inner_name)
if not resolved:
return {"cover_photo_url": None, "video_url": None, "docs": []}
manifest = _load_mirror_manifest(marketplace_id)
cover_url: Optional[str] = None
cover_ref = resolved.get("cover_photo_ref")
if isinstance(cover_ref, tuple):
ref_kind, target = cover_ref
if ref_kind == "internal":
local_path = repo_root / target
if local_path.is_file():
cover_url = (
f"/api/marketplace/curated/{marketplace_id}/{plugin_name}"
f"/asset/{target}"
)
elif ref_kind == "external":
cover_url = _resolve_external_via_mirror(
marketplace_id, plugin_name, target, manifest,
)
docs: List[DocEntry] = []
for link in resolved.get("doc_links") or []:
if not hasattr(link, "kind"):
continue
if link.kind == "internal":
local_path = repo_root / link.path
if not local_path.is_file():
continue
docs.append(DocEntry(
name=link.name,
url=(
f"/api/marketplace/curated/{marketplace_id}/{plugin_name}"
f"/doc/{link.path}"
),
))
else:
served = _resolve_external_via_mirror(
marketplace_id, plugin_name, link.url, manifest,
)
if served is None:
continue
docs.append(DocEntry(name=link.name, url=served))
from app.markdown_render import render_safe
out: Dict[str, Any] = {
"cover_photo_url": cover_url,
"video_url": resolved.get("video_url"),
"docs": docs,
}
# Rich user-facing fields — same shape as the plugin-level enrichment.
# Presence check (``in``) rather than truthy check so the resolver's
# contract is respected: if the resolver decided to include a key,
# the API propagates it. Future falsy-but-valid fields (e.g. a bool
# ``featured: false`` or numeric ``priority: 0``) inherit correctly
# instead of silently falling back to the parent — the trap @cvrysanek
# flagged in the multi-persona review. String fields stay truthy-
# guarded by the resolver itself, so today's behaviour is unchanged.
if "display_name" in resolved:
out["display_name"] = resolved["display_name"]
if "tagline" in resolved:
out["tagline"] = resolved["tagline"]
# Per-item category override — caller merges with the parent plugin's
# category, preferring the override when set.
if "category" in resolved:
out["category"] = resolved["category"]
if "description" in resolved:
out["description_long_html"] = render_safe(resolved["description"])
if "when_to_use" in resolved:
out["when_to_use_html"] = render_safe(resolved["when_to_use"])
if "invocation" in resolved:
out["invocation"] = resolved["invocation"]
use_cases_raw = resolved.get("use_cases") or []
if use_cases_raw:
cards = [_safe_use_case(uc) for uc in use_cases_raw]
cards = [c for c in cards if c is not None]
if cards:
out["use_cases"] = cards
sample = _safe_sample_interaction(resolved.get("sample_interaction"))
if sample is not None:
out["sample_interaction"] = sample
return out
# Module-level cache for marketplace-metadata.json reads. Keyed by
# ``(marketplace_id, mtime_ns)`` so a curator's `git push` (which updates
# the file's mtime after the next sync's `git pull`) implicitly invalidates
# the cached parse. The 24-plugin listing endpoint reads from the same
# marketplace's metadata 24 times per page; without this cache that's 24
# disk reads + JSON parses per request. With cache: 1.
#
# OrderedDict + popitem(last=False) on overflow gives us a bounded LRU
# without a third-party dependency. Earlier versions used a per-marketplace
# eviction predicate that only swept stale entries for the CURRENT marketplace
# — at N>100 distinct marketplaces the inner predicate matched zero entries,
# so the cap silently failed and memory grew linearly. The bounded LRU here
# guarantees the dict size never exceeds _PLUGIN_METADATA_CACHE_MAX.
_PLUGIN_METADATA_CACHE_MAX = 256
_PLUGIN_METADATA_CACHE: "OrderedDict[Tuple[str, int], Dict[str, Any]]" = OrderedDict()
def _read_metadata_cached(marketplace_id: str) -> Dict[str, Any]:
"""Return the parsed marketplace-metadata.json for a given marketplace.
Cached by mtime so curator edits land at next request without explicit
invalidation. Missing / malformed files degrade to ``{}`` (delegated to
:func:`src.marketplace_metadata.read_marketplace_metadata` which already
swallows OSError / JSONDecodeError). Cache is bounded LRU; the oldest
entry is dropped when the cap is reached.
"""
from src.marketplace_metadata import (
MARKETPLACE_METADATA_REL,
read_marketplace_metadata,
)
repo_root = Path(get_marketplaces_dir()) / marketplace_id
metadata_path = repo_root / MARKETPLACE_METADATA_REL
try:
mtime_ns = metadata_path.stat().st_mtime_ns
except OSError:
# File doesn't exist or unreadable — read_marketplace_metadata handles
# the same case below; we just skip the cache lookup.
return read_marketplace_metadata(repo_root)
key = (marketplace_id, mtime_ns)
cached = _PLUGIN_METADATA_CACHE.get(key)
if cached is not None:
# Touch the entry so the LRU bookkeeping treats it as recently used.
_PLUGIN_METADATA_CACHE.move_to_end(key)
return cached
parsed = read_marketplace_metadata(repo_root)
_PLUGIN_METADATA_CACHE[key] = parsed
# Bounded LRU: drop oldest entries until we're back under the cap.
while len(_PLUGIN_METADATA_CACHE) > _PLUGIN_METADATA_CACHE_MAX:
_PLUGIN_METADATA_CACHE.popitem(last=False)
return parsed
def _curated_plugin_enrichment(
marketplace_id: str,
plugin_name: str,
) -> Dict[str, Any]:
"""Load plugin-level rich content from marketplace-metadata.json.
Returns a dict shaped for direct merge into ``PluginDetailResponse``:
``display_name``, ``tagline``, ``description_long_html``, ``use_cases``,
``sample_interaction`` — any subset of those keys may be missing when
the curator hasn't filled the corresponding field.
Markdown is rendered to safe HTML here (via
:func:`app.markdown_render.render_safe`) so the template can inject the
body with ``{{ x | safe }}`` without a client-side markdown library.
This is the plugin-level analogue of ``_curated_inner_enrichment`` — both
read on demand from the working tree so curator edits don't need a full
sync cycle to land in the UI. The other persisted plugin-level fields
(cover_photo_url, video_url, category, doc_links) continue to flow via
``marketplace_plugins`` rows written at sync time — only the NEW rich
fields are read on-demand here.
"""
from app.markdown_render import render_safe
from src.marketplace_metadata import resolve_plugin_metadata
metadata = _read_metadata_cached(marketplace_id)
resolved = resolve_plugin_metadata(metadata, plugin_name)
if not resolved:
return {}
out: Dict[str, Any] = {}
# Presence check (`in`) so future falsy-but-valid resolver fields
# (bool / int / "" with explicit-empty semantics) survive — see the
# parallel comment in `_curated_inner_enrichment` for the trap rationale.
if "display_name" in resolved:
out["display_name"] = resolved["display_name"]
if "tagline" in resolved:
out["tagline"] = resolved["tagline"]
if "description" in resolved:
out["description_long_html"] = render_safe(resolved["description"])
use_cases_raw = resolved.get("use_cases") or []
if use_cases_raw:
cards = [_safe_use_case(uc) for uc in use_cases_raw]
cards = [c for c in cards if c is not None]
if cards:
out["use_cases"] = cards
sample = _safe_sample_interaction(resolved.get("sample_interaction"))
if sample is not None:
out["sample_interaction"] = sample
return out
def _curated_inner_cover(
marketplace_id: str,
plugin_name: str,
kind: str,
inner_name: str,
manifest: Optional[Dict[Tuple[str, str], Any]] = None,
metadata: Optional[Dict[str, Any]] = None,
) -> Optional[str]:
"""Cheap helper: just the cover URL for one inner item.
Used by ``curated_detail`` to populate ``InnerItemSummary.cover_photo_url``
on the parent-plugin's skills/agents card list. Called once per inner
item; metadata + manifest are loaded once per plugin and passed in to
avoid N disk reads.
"""
from src.marketplace_metadata import resolve_inner_metadata
if metadata is None:
# Route through the mtime cache; this helper gets called once per
# inner item on the parent-plugin's skills/agents card list, so
# cache miss only happens on the first card.
metadata = _read_metadata_cached(marketplace_id)
if manifest is None:
manifest = _load_mirror_manifest(marketplace_id)
section_kind = "skills" if kind == "skill" else "agents"
resolved = resolve_inner_metadata(metadata, plugin_name, section_kind, inner_name)
cover_ref = resolved.get("cover_photo_ref") if resolved else None
if not isinstance(cover_ref, tuple):
return None
ref_kind, target = cover_ref
if ref_kind == "internal":
local_path = (Path(get_marketplaces_dir()) / marketplace_id / target)
if local_path.is_file():
return (
f"/api/marketplace/curated/{marketplace_id}/{plugin_name}"
f"/asset/{target}"
)
return None
return _resolve_external_via_mirror(
marketplace_id, plugin_name, target, manifest,
)
@router.get(
"/curated/{marketplace_id}/{plugin_name}/skill/{skill_name}",
response_model=InnerDetailResponse,
)
async def curated_skill_detail(
marketplace_id: str,
plugin_name: str,
skill_name: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
plugin_root = (
Path(get_marketplaces_dir()) / marketplace_id / "plugins" / plugin_name
)
res = _read_inner(plugin_root, "skills", skill_name, is_dir_layout=True)
skill_dir = _safe_join(plugin_root, "skills", skill_name)
if res is None or skill_dir is None:
raise HTTPException(status_code=404, detail="skill_not_found")
text, relpath = res
fm = _parse_frontmatter(text)
parent = _curated_inner_parent_fields(conn, marketplace_id, plugin_name)
enrichment = _curated_inner_enrichment(
marketplace_id, plugin_name, "skill", skill_name,
)
# Merge parent fallback fields with curator overrides BEFORE unpacking
# — Python function-call `**a, **b` with overlapping keys raises
# TypeError, it doesn't merge like a literal dict does. Today only
# `category` overlaps (both parent + enrichment may set it), but the
# explicit merge keeps the unpack future-proof against any new field
# added to both layers.
merged = {**parent, **enrichment}
return InnerDetailResponse(
marketplace_id=marketplace_id,
plugin_name=plugin_name,
kind="skill",
name=fm.get("name") or skill_name,
description=fm.get("description"),
body=_frontmatter_body(text),
relpath=relpath,
bundle_size=_bundle_size(skill_dir),
files=_walk_files(skill_dir),
**merged,
)
@router.get(
"/curated/{marketplace_id}/{plugin_name}/agent/{agent_name}",
response_model=InnerDetailResponse,
)
async def curated_agent_detail(
marketplace_id: str,
plugin_name: str,
agent_name: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
conn: duckdb.DuckDBPyConnection = Depends(_get_db),
):
plugin_root = (
Path(get_marketplaces_dir()) / marketplace_id / "plugins" / plugin_name
)
res = _read_inner(plugin_root, "agents", agent_name, is_dir_layout=False)
agent_path = _safe_join(plugin_root, "agents", f"{agent_name}.md")
if res is None or agent_path is None:
raise HTTPException(status_code=404, detail="agent_not_found")
text, relpath = res
fm = _parse_frontmatter(text)
# Agents are flat single-file .md — bundle = file size, files = one entry.
try:
agent_size = agent_path.stat().st_size
except OSError:
agent_size = 0
parent = _curated_inner_parent_fields(conn, marketplace_id, plugin_name)
enrichment = _curated_inner_enrichment(
marketplace_id, plugin_name, "agent", agent_name,
)
# See curated_skill_detail above — explicit merge avoids the
# TypeError that `**parent, **enrichment` raises when both supply
# an overlapping key (e.g. `category`).
merged = {**parent, **enrichment}
return InnerDetailResponse(
marketplace_id=marketplace_id,
plugin_name=plugin_name,
kind="agent",
name=fm.get("name") or agent_name,
description=fm.get("description"),
body=_frontmatter_body(text),
relpath=relpath,
bundle_size=agent_size,
files=[FileEntry(path=f"{agent_name}.md", size=agent_size)],
**merged,
)
# ---------------------------------------------------------------------------
# Asset / doc / mirrored serving endpoints (v32)
# ---------------------------------------------------------------------------
#
# Three sibling endpoints that serve the binary content referenced from
# `marketplace-metadata.json`. All three:
#
# * are gated by `require_resource_access(MARKETPLACE_PLUGIN, "{mp}/{plugin}")`
# so a user without RBAC can't side-load assets even with a direct URL,
# * resolve a candidate path with `Path.resolve(strict=True)` and verify the
# result lives under the expected root via `is_relative_to()` — defense
# against `..` / absolute paths / symlinks pointing out of the tree,
# * use FastAPI's `FileResponse` so Content-Type detection comes from the
# stdlib mimetypes module (good enough for the allowlisted set; binary
# fallback for anything we don't recognize).
def _doc_disposition(filename: str) -> dict:
"""Force-download headers for the /doc and /mirrored doc paths.
Browsers honor the frontend's `download` attribute only for same-origin
URLs; the explicit Content-Disposition header makes the download
behavior reliable regardless of how the link was opened (right-click →
open, programmatic fetch, curl, …). Filename is the basename so the
browser save dialog shows something readable.
"""
safe = Path(filename).name or "download"
return {"Content-Disposition": f'attachment; filename="{safe}"'}
# Mapping from validated image extension to the Content-Type we serve.
# Pinned (not stdlib mimetypes) so an unexpected file with a known extension
# can't push us into ``text/html`` territory — this endpoint NEVER serves
# anything but image bytes labelled as image bytes.
_ASSET_CONTENT_TYPE = {
".png": "image/png",
".jpg": "image/jpeg",
".jpeg": "image/jpeg",
".webp": "image/webp",
}
# Defense-in-depth headers applied to every /asset/ response. ``nosniff``
# stops browsers from second-guessing our Content-Type; the strict CSP is
# a belt-and-suspenders block — even if a future regression let HTML
# through, the browser still won't execute scripts/iframes/etc.
_ASSET_SECURITY_HEADERS = {
"X-Content-Type-Options": "nosniff",
"Content-Security-Policy": "default-src 'none'; img-src 'self'; style-src 'unsafe-inline'",
}
@router.get("/curated/{marketplace_id}/{plugin_name}/asset/{path:path}")
async def curated_asset(
marketplace_id: str,
plugin_name: str,
path: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
):
"""Serve an internal image asset from the cloned marketplace working tree.
Paths are repo-root-relative — ``{path}`` may be e.g.
``.agnes/cover.png`` or ``plugins/foo/icon.png``.
**Image-only by contract.** The endpoint is the source of cover photos
referenced from ``marketplace-metadata.json`` and from inner skill / agent
cards. A curator who could land an arbitrary file in the cloned repo
(HTML, JS, SVG with inline ``<script>``) would otherwise have a
same-origin XSS via this endpoint, since the response shares the
cookie scope with ``/admin`` and ``/api/me/*``. Three layered checks:
1. Extension must be in :data:`src.marketplace_asset_validation.IMAGE_EXTENSIONS`
(``.png``/``.jpg``/``.jpeg``/``.webp``); anything else → 415.
2. Body must pass :func:`src.marketplace_asset_validation.validate_image_file`
magic-bytes check; mismatch → 415 (defeats the rename-extension
attack: ``evil.png`` carrying ``<script>`` bytes).
3. ``Content-Type`` is pinned from the extension table above (not
stdlib mimetypes), so the response is never served as ``text/html``
even if mimetypes were misconfigured.
SVG is intentionally not in the allowlist — ``<script>`` inside SVG
executes in the browser. ``X-Content-Type-Options: nosniff`` plus a
strict CSP harden the response further.
Inline rendering (no ``Content-Disposition``) — covers display in
``<img>``, not as a download.
"""
from src.marketplace_asset_validation import IMAGE_EXTENSIONS, validate_image_file
repo_root = Path(get_marketplaces_dir()) / marketplace_id
if not repo_root.exists():
raise HTTPException(status_code=404, detail="marketplace_not_synced")
safe = _safe_join(repo_root, path)
if safe is None or not safe.is_file():
raise HTTPException(status_code=404, detail="asset_not_found")
ext = safe.suffix.lower()
if ext not in IMAGE_EXTENSIONS:
raise HTTPException(
status_code=415,
detail=f"unsupported_asset_extension: {ext or '(none)'}",
)
try:
body = safe.read_bytes()
except OSError:
raise HTTPException(status_code=404, detail="asset_not_readable")
validation = validate_image_file(safe.name, body)
if not validation.ok:
raise HTTPException(
status_code=415,
detail=f"asset_validation_failed: {validation.reason}",
)
return FileResponse(
safe,
media_type=_ASSET_CONTENT_TYPE[ext],
headers=_ASSET_SECURITY_HEADERS,
)
@router.get("/curated/{marketplace_id}/{plugin_name}/doc/{path:path}")
async def curated_doc(
marketplace_id: str,
plugin_name: str,
path: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
):
"""Serve an internal doc path from the cloned marketplace working tree.
Same path-traversal guard as ``/asset/``. Adds an allowlist check — the
doc endpoint refuses to serve a file whose extension isn't in the
documented PDF / Markdown / plain text set (HTTP 415). Defense-in-depth
even though the marketplace-metadata parser already rejects out-of-allowlist
extensions during the doc_link parse — a curator who edits the working
tree directly (or whose JSON survived parsing because of a generic
extension match elsewhere) shouldn't be able to land a .docx through
a re-served doc URL.
Force-download via Content-Disposition: attachment — clicking a doc
link in the UI saves the file to disk rather than opening it in a tab.
"""
from src.marketplace_asset_validation import DOC_EXTENSIONS
repo_root = Path(get_marketplaces_dir()) / marketplace_id
if not repo_root.exists():
raise HTTPException(status_code=404, detail="marketplace_not_synced")
safe = _safe_join(repo_root, path)
if safe is None or not safe.is_file():
raise HTTPException(status_code=404, detail="doc_not_found")
if safe.suffix.lower() not in DOC_EXTENSIONS:
raise HTTPException(
status_code=415,
detail=f"unsupported_doc_extension: {safe.suffix.lower() or '(none)'}",
)
return FileResponse(safe, headers=_doc_disposition(safe.name))
@router.get("/curated/{marketplace_id}/{plugin_name}/mirrored/{key:path}")
async def curated_mirrored(
marketplace_id: str,
plugin_name: str,
key: str,
_user: dict = Depends(require_resource_access(
ResourceType.MARKETPLACE_PLUGIN, "{marketplace_id}/{plugin_name}",
)),
):
"""Serve a mirrored external asset from the marketplace cache.
``key`` is the ``local`` relpath stored in the cache manifest minus the
leading ``<plugin>/`` prefix — the mirror writes files at
``${DATA_DIR}/marketplace-cache/<slug>/<plugin>/<rest>``, and this
endpoint expects ``{plugin}/{rest}`` so the same path-resolution check
catches escapes whether the caller provided ``cover.png`` or
``docs/abc-setup.pdf``.
Doc-shaped paths (key starting with ``docs/``) get the same force-download
treatment as the internal /doc/ endpoint. Cover photos under the cache
root render inline so the <img> tag works.
"""
cache_root = get_marketplace_cache_dir() / marketplace_id / plugin_name
if not cache_root.exists():
raise HTTPException(status_code=404, detail="mirror_cache_missing")
safe = _safe_join(cache_root, key)
if safe is None or not safe.is_file():
raise HTTPException(status_code=404, detail="mirrored_asset_not_found")
if key.startswith("docs/"):
return FileResponse(safe, headers=_doc_disposition(safe.name))
return FileResponse(safe)