anthropics/claude-plugins-official
1
0
Fork 0
mirror of https://github.com/anthropics/claude-plugins-official.git synced 2026-06-13 22:26:03 -03:00
Code Issues Packages Projects Releases Wiki Activity
453 Commits 390 Branches 0 Tags
Commit Graph

6 Commits

Author SHA1 Message Date
Mohamed Hegazy
475038edfc
security-guidance: emit HTTP error codes + fix sdk_bootstrap phase/err encoding 
Fills two failure-visibility gaps in plugin telemetry.

## Gap 1: HTTP errors from _call_claude invisible

Before: a 4xx/5xx response from the LLM API caused `_call_claude` to
return None and produce ZERO fingerprint in tengu_hook_plugin_metrics.
A failed call looked identical to "no review needed". The recent
deprecation-400 outage (PR #2105, output_format → output_config.format,
#2098) was invisible in aggregate dashboards until a user manually
reported errors from their debug log. Cohort-specific or partial
outages would never show up in BQ.

Fix: add `http_err_last` (most recent status) and `http_err_count`
to the existing `_USAGE` accumulator in `_base.py`. `_usage_metrics()`
snapshots them whenever count > 0 (skip-path no-pollute contract
preserved when count == 0). All `_call_claude` error sites now call
the new `_record_http_error()` helper alongside the existing
`_last_call_claude_http_error` module-state assignment.

Now any future API failure category is queryable in BQ in real time:

  SELECT
    DATE(server_timestamp, "America/Los_Angeles") AS d,
    CAST(JSON_VALUE(additional_metadata, "$.http_err_last") AS INT64) AS code,
    COUNT(*) AS n
  FROM ... WHERE event_name = "tengu_hook_plugin_metrics"
    AND JSON_VALUE(additional_metadata, "$.pluginId") LIKE "%security-guidance%"
    AND JSON_VALUE(additional_metadata, "$.http_err_count") IS NOT NULL
  GROUP BY d, code ORDER BY d, n DESC

## Gap 2: sdk_bootstrap_phase / sdk_bootstrap_err always NULL in BQ

Before: ensure_agent_sdk.py emitted these as strings (e.g. "pip",
"dns_fail"). CC's plugin-metrics pipeline silently drops
plugin-emitted string values — only bool|finite-number plugin metrics
reach BigQuery. (CC-core fields like `subscription_type` are exempt
because they're injected downstream of plugin validation.)

Confirmed empirically: ~185K BUILD_FAILED rows in BQ over the past 2
days had `sdk_bootstrap_phase` = NULL and `sdk_bootstrap_err` = NULL
despite the Python code emitting them. ~28K BUILD_FAILED sessions/day
had no diagnostic split — flying blind on whether the failures are
pip-no-match vs dns-fail vs ssl-verify vs proxy-auth etc.

Fix: encode phase + err_kind as stable integers via
SDK_BOOTSTRAP_PHASE_CODES and SDK_BOOTSTRAP_ERR_CODES. Phase: 1=pre,
2=venv, 3=pip, 4=main. Err: 10 known categories (1-10), 11-98
reserved, 99 = uncategorized catch-all (covers "exc:<X>", "other:<X>",
and unmapped strings). APPEND-ONLY for telemetry stability.

Also corrects the misleading "CC accepts string metric values"
comment in ensure_agent_sdk.py that led to the bug originally.

Verified locally on macOS Python 3.13:

  - py_compile clean.
  - 32 new tests in test_telemetry_failure_signals.py (added to
    internal test suite at sg-staging/tests/, not in this PR):

      * 4 HTTP-error tracking unit tests: _record_http_error increments
        count + tracks most-recent; handles None/invalid; -1 for
        network/timeout.
      * 4 _usage_metrics emission tests: empty when no activity;
        successful call has no http_err fields; failure-only has http_err
        and no api_calls; mixed has both.
      * 1 contract test: every emitted value is bool|finite-number
        (catches future regression of the string-dropping bug class).
      * 13 sdk_bootstrap encoding tests (parametrized over the 10 known
        err_kind categories + 5 catch-all shapes): each maps to the
        right integer; unknown phase = 0; unknown err = 99.
      * 1 static-shape regression catcher: every `err_kind = "..."`
        string in ensure_agent_sdk.main() must be in
        SDK_BOOTSTRAP_ERR_CODES (otherwise new err_kinds silently
        collapse to 99).
      * 2 emit-shape regression catchers: the assignments in main() go
        through _encode_phase / _encode_err_kind helpers (no raw
        strings); no literal string values for sdk_bootstrap_phase/err.
      * 1 comment-accuracy: the misleading "CC accepts string metric
        values" comment is gone.

  - Full suite: 437/437 pass + 2 skipped (live API tests, opt-in).

NOT verified end-to-end against BQ — would require shipping +
observing in production for 24h to confirm the http_err and
sdk_bootstrap_phase/err fields actually appear in
tengu_hook_plugin_metrics rows. The unit tests pin the contract; if
the wire shape is broken, BQ will show NULL for the new fields and we
revisit (with the same diagnostic the BUILD_FAILED bug gave us).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-31 08:34:35 -07:00
Mohamed Hegazy
84011d43b1
security-guidance: migrate from deprecated output_format to output_config.format (#2098) 
Fixes #2098. The Anthropic Messages API moved structured-output
schema specification from a top-level `output_format` field to a
nested `output_config.format` field, per
https://platform.claude.com/docs/en/build-with-claude/structured-outputs.

Per docs the old form "will continue working for a transition period"
— and indeed for api-key + non-streaming auth it still returns HTTP
200 (verified via live API). But OAuth Bearer users with CLI 2.1.158
hit `invalid_request_error: output_format: This field is deprecated.
Use 'output_config.format' instead.` consistently — reporter saw 462
errors in one day. The trigger appears to be auth mode + possibly
stream:true (their controlled curl bypass used Bearer + stream=true);
api-key + non-streaming was my initial repro attempt and didn't fire.

The bug only affected `_call_claude` (the legacy direct-urllib path).
The agentic `_agentic_review` path goes through claude_agent_sdk →
subprocesses to the `claude` CLI binary, which already uses the new
`output_config.format` shape correctly (per src/utils/sideQuery.ts:263
in claude-cli-internal). So this PR only needs to fix the plugin's
direct HTTP path.

This commit:

1. llm.py: rewrite the payload literal in `_call_claude` to use
   `output_config: { format: { type: 'json_schema', schema: ... } }`
   instead of top-level `output_format`.

2. llm.py: in the adaptive-thinking branch, MERGE `effort: "high"`
   into the existing `output_config` dict instead of reassigning.
   Reassignment would silently clobber the format schema set in (1).
   The pre-existing code did `payload["output_config"] = {"effort":
   "high"}` which was correct WHEN output_format was top-level (and
   output_config wasn't otherwise used). With the migration the
   existing dict carries the schema, so we extend it not replace it.

Verified locally on macOS Python 3.13:

  - py_compile clean.
  - Existing 401 tests still pass — 0 regression.
  - 6 new tests in test_2098_output_config_format.py (added to
    internal test suite at sg-staging/tests/, not in this PR):

      * 2 static-shape: the `_call_claude` source no longer contains
        top-level `"output_format":` AND uses `output_config`. The
        adaptive-thinking branch does NOT reassign output_config (and
        DOES set output_config['effort']). Catches the regression
        class where a future refactor reintroduces either bug.
      * 2 payload-shape unit (mocked urllib): both thinking_budget=0
        and thinking_budget>0+adaptive code paths produce a payload
        with the correct `output_config.format` shape AND no
        `output_format` top-level. The adaptive path verifies both
        `format` and `effort` coexist in output_config (i.e., the
        merge fix works).
      * 2 live-API gating (skip-on-no-key): the new shape returns
        HTTP 200 against api.anthropic.com; the old shape's current
        status is recorded for canary purposes (still 200 for
        api-key today, but reporter shows it's 400 for OAuth).

  - Full suite: 405/405 pass + 2 skipped (live API tests, opt-in).
  - The reporter's exact deprecation 400 message reproduces if you
    swap auth to OAuth Bearer + stream:true (could not test locally
    without extracting the keychain OAuth token, which was out of
    scope). The fix shape is API-contract-level so it doesn't depend
    on which auth mode triggers the 400.

NOT verified end-to-end via OAuth-authenticated plugin invocation on
my machine (auto-mode classifier correctly declined to extract the
keychain token). Reporter's 462 production errors + the docs
migration notice + the live-API HTTP 200 on the new form are
sufficient evidence to ship.

Closes #2098.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-30 20:11:41 -07:00
Mohamed Hegazy
0d22ba3501
security-guidance: respect CLAUDE_CONFIG_DIR for plugin state files (#1868) 
Fixes #1868 — when CLAUDE_CONFIG_DIR is set to a non-default location
(e.g. ~/.config/claude for XDG compliance, or a multi-tenant install
path), the plugin still wrote state files to the hardcoded ~/.claude/
path, leaving stale state and breaking CLAUDE_CONFIG_DIR's purpose.

Resolution precedence (highest first):
  1. SECURITY_WARNINGS_STATE_DIR  — plugin-specific override (existing)
  2. CLAUDE_CONFIG_DIR/security    — CC's config-dir env (new — #1868)
  3. ~/.claude/security            — default fallback (unchanged)

Empty-string env vars (e.g. CLAUDE_CONFIG_DIR= in a misconfigured
shell) are treated as not-set so the empty path doesn't collide with
os.path.join and silently write to /security at the filesystem root.

Implementation: a single state_dir() helper in _base.py is the source
of truth for resolution. All five modules that previously had inline
SECURITY_WARNINGS_STATE_DIR / ~/.claude/security resolutions
(_base.py, session_state.py, ensure_agent_sdk.py, llm.py, and one
site in security_reminder_hook.py) now call state_dir() instead.
Re-implementing the precedence inline risks drift — one module gets
a future fix, others don't.

The helper is called per-invocation rather than cached at import time
so test monkeypatches of the env vars take effect, and so a long-
running test or future shared-process scenario can change the env
between calls and have the next call observe the new value. The
per-call cost is negligible compared to the subprocess-spawn cost
the hooks pay every fire in production.

Three hardcoded ~/.claude/security strings remain but are NOT
functional resolutions:
  - _base.py:39: the fallback BRANCH inside state_dir() itself
  - ensure_agent_sdk.py:6, :11: docstring text describing default
                                location for users

Verified locally on macOS Python 3.13:

  - py_compile clean on all 5 modified files.
  - Existing 45 smoke + extensibility tests still pass.
  - 14 new tests in test_claude_config_dir.py (added to internal test
    suite at sg-staging/tests/, not in this PR):

      * 7 resolution-semantics: default fallback, CLAUDE_CONFIG_DIR
        override, SECURITY_WARNINGS_STATE_DIR beats both, tilde
        expansion, empty-string handling (CLAUDE_CONFIG_DIR= must
        fall back, NOT join to /security).
      * 4 static-shape: each of session_state / ensure_agent_sdk /
        llm / security_reminder_hook either imports state_dir from
        _base OR has zero resolution patterns. Catches the
        regression where someone adds a new state-file writer and
        re-implements resolution inline, missing the
        CLAUDE_CONFIG_DIR branch.
      * 3 end-to-end: with CLAUDE_CONFIG_DIR set, get_state_file /
        get_lock_file return paths under <CLAUDE_CONFIG_DIR>/security/;
        save_state round-trip writes a file to the redirected path
        and re-reads the same contents.

  - 59/59 pass total (45 existing + 14 new) in 2.54s.

NOT verified end-to-end with a real CC instance setting
CLAUDE_CONFIG_DIR. The shape tests catch the regression class
(hardcoded ~/.claude/), and the end-to-end test pins the behavior
that user state files actually land at the redirected path.

Closes #1868.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-28 23:57:10 -07:00
Mohamed Hegazy
c11244778d
Address Windows verification: --prefer-binary + pywin32 bootstrap 
The first round of this PR removed SKIP_WIN32, fixed venv_py to use
Scripts/python.exe, and added Lib/site-packages to the consumer glob —
all necessary. Windows verification (Win11 ARM64, Py 3.13, Git Bash)
showed two more blockers, both addressed here.

1. Pip dependency resolver picks unbuildable cryptography on ARM64.

   Without --prefer-binary, pip picks a cryptography version with no
   published ARM64 wheel and tries to build it from source. That needs
   Rust/Cargo, almost never present on user machines → BUILD_FAILED
   with err_kind=other:cryptography. A binary wheel exists for an
   adjacent version (cryptography-46.0.3-cp311-abi3-win_arm64.whl);
   --prefer-binary tells pip to pick it. Cross-platform safe (no-op
   where the latest version already has a wheel).

2. pywin32 .pth files aren't processed by sys.path.insert().

   With the venv built, ensure_agent_sdk.py's post-build probe passes
   (it runs from venv_py, where Python's site.py at startup processes
   pywin32.pth and registers win32/, win32/lib/ plus runs
   pywin32_bootstrap.py to set the DLL search dir). But llm.py runs in
   the hook's SYSTEM Python and adds the venv via sys.path.insert(),
   which doesn't trigger site.py at all. Without the bootstrap, the
   SDK's mcp.client.stdio → mcp.os.win32.utilities chain raises
   ModuleNotFoundError: pywintypes and the agentic reviewer falls back
   to single-shot silently — exactly the symptom this PR is trying to
   fix. The probe says NOOP_VENV; the actual consumer fails. Probe and
   consumer use different Pythons.

   Replicate what site.py would do: after inserting site-packages,
   also insert win32/ and win32/lib/, then exec pywin32_bootstrap.py.
   Pulled into a shared helper _inject_agent_sdk_venv_into_syspath()
   so both consumer sites (3P SDK fallback, agentic_review fallback)
   call the same code — Windows handling stays in one place.

Verified on macOS (POSIX path unchanged):
- Helper end-to-end test: POSIX-layout venv detected + fake package
  imports successfully via the injected path
- Windows-layout venv also detected; win32 branch correctly skipped
  via sys.platform check
- Both files pass py_compile

Credit: @mhegazy verified the previous commit on Win11 ARM64 / Py 3.13
/ Git Bash, surfaced both issues end-to-end, and provided the exact
fix patterns. This commit applies them with the pywin32 part factored
into a shared helper (vs. inlining at both consumer sites).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-27 15:07:33 -07:00
Mohamed Hegazy
4decd2e3b2
Enable agentic commit reviewer on Windows 
The agentic reviewer is silently no-op on Windows today. SessionStart
bootstrap (ensure_agent_sdk.py) short-circuits with SKIP_WIN32 because
the consumer glob in llm.py only matches POSIX venv layout
(lib/pythonX.Y/site-packages). On Windows, venvs use Lib/site-packages
(capital L, no pythonX.Y subdir), so even if a venv got built the
glob wouldn't find its contents.

Result: Windows users on default installs (no system-wide
claude_agent_sdk) get layer 1 (pattern warnings) and layer 2 (single-
shot LLM diff review) but not layer 3 — the cross-file agentic review
that catches IDOR, auth-bypass, cross-file SSRF, and other things that
need to read related files. Plugin description claims layer 3 but it
silently doesn't run.

Three changes:

1. llm.py — extend the consumer glob (2 sites: 3P SDK fallback at
   ~L297, agentic_review fallback at ~L1090) to also match the Windows
   Lib/site-packages layout, so a venv built on Windows is actually
   discoverable.

2. ensure_agent_sdk.py — remove the sys.platform == 'win32' early-exit
   so the SessionStart bootstrap builds the venv on Windows too.
   Outcome code 4 (formerly SKIP_WIN32) is retired but not reused so
   pre-fix telemetry rows still decode correctly.

3. ensure_agent_sdk.py — venv_py path now branches on sys.platform:
   Windows venvs put the interpreter at Scripts\python.exe; POSIX
   uses bin/python. Previously assumed POSIX, so even with the glob
   fix, the post-build SDK-importability probe would fail on Windows.

Verified locally on macOS:
- glob test: both layouts now match (POSIX venv detected, simulated
  Windows venv also detected via the new Lib/site-packages branch)
- both files pass py_compile
- POSIX path unchanged (sys.platform != 'win32' so old branch runs)

Not verified on Windows in this commit — needs an actual Windows
runner to confirm the venv build + SDK import + subprocess plumbing
all work end-to-end. The SDK spawns a child claude.exe; Windows
process plumbing has its own quirks (shell semantics, path escaping)
that may surface separately. Worth a controlled rollout (one-week
soak under env-var opt-in before flipping default).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-27 14:50:51 -07:00
Mohamed Hegazy
0bde168648
Update security-guidance plugin 2026-05-26 14:06:52 -07:00