Resolves conflict in .github/workflows/pytest.yml. Both branches added
--ignore-vuln CVE-2026-3219 to pip-audit (this branch in aee5278, main
in #1779). Took main's version since #1779 has the canonical, more
detailed rationale comment block (links to pypa/pip#13870 fix PR,
explains the local-only attack surface and why it does not apply to CI).

Also addresses the latest Copilot review comments on this PR:

- tests/eval/test_eval_prompt_change.py: scope the sys.path mutation
  to module load only and restore it in a finally block. eval-prompt-
  change.py imports sibling modules (_anthropic_api, _eval_common) by
  bare name, so EVAL_DIR must be on sys.path during exec_module();
  removing the entry afterwards prevents the addition from leaking
  into other tests' import resolution. Tests still pass: 85 in this
  file, 107 with workspace-limits tests included.

- .agents/sessions/2026-04-25-session-1714.json: rewrite workLog steps
  1 and 7 evidence so they no longer reference .agents/audit/eval-1741/
  or .agents/audit/eval-1755/ paths that do not exist in the repo (the
  artifacts were transient and dropped in 31ed283). Step 1 now points
  to the reproducer command; step 7 points to the per-agent table in
  the PR Honesty notes section.

Pulls in main's PR #1779 (pip CVE ignore w/ rationale) and PR #1762
(enterprise-patterns rule) as part of the merge.

Refs #1755

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

https://claude.ai/code/session_01XALNMgBrrH7Ya8dM1CjYzo

The session log used legacy field names (serenaInit, lintRun, commitCreated,
etc.) that the current validate_session_json.py rejects. Renamed to the
required schema fields: sessionStart adds serenaActivated, serenaInstructions,
branchVerified, notOnMain; sessionEnd renames logComplete -> checklistComplete,
lintRun -> markdownLintRun, commitCreated -> changesCommitted, jsonValid ->
validationPassed, and adds handoffPreserved + serenaMemoryUpdated.

Hardened scripts/validate_pr_review_config.py against CWE-22 path traversal
by routing the user-supplied config_path through validate_safe_path against
_PROJECT_ROOT before any open(). Matches the pattern already used in
scripts/validate_session_json.py. Added 3 CLI tests covering relative
traversal, absolute paths outside the repo, and the default-path happy path.
Removed unused pytest/Path imports flagged by ruff.

Python Security Checks (pip-audit CVE-2026-3219) is resolved by the rebase
on origin/main, which now ignores the CVE per workflow update in #1779.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…iteration limits (#1671)

* fix(commands): bound pr-review and context-hub-setup with output and iteration limits

Adds invocation_limits and output_constraints to pr-review-config.yaml:
- all-open caps at 5 PRs per invocation with overflow reporting
- completion gate caps at 3 retry iterations before escalation
- per-PR agent output caps at 120 lines with session-file persistence
- summary uses required tabular columns

Updates pr-review.md Step 1, Completion Gate, and Step 6 to reference the
new config keys so agents enforce the limits at runtime.

Adds Completion Criteria, Task Budget, and Output Constraints sections to
context-hub-setup.md so setup has a verifiable done-when condition and
bounded retries.

Fixes #1668

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(commands): address PR review feedback for output constraints (#1668)

- Fix retry contradiction in context-hub-setup.md completion criteria
- Fix {pr} placeholder to {number} for consistency in config
- Clarify Step 3 output rules to allow Commands available block
- Clarify retries vs iterations semantics in completion gate
- Remove inline defaults from pr-review.md, single-source from config
- Add invocation_limits and output_constraints validation to schema
- Add tests for new validation rules

* fix(sessions): repair session log validation; address security findings

The session log used legacy field names (serenaInit, lintRun, commitCreated,
etc.) that the current validate_session_json.py rejects. Renamed to the
required schema fields: sessionStart adds serenaActivated, serenaInstructions,
branchVerified, notOnMain; sessionEnd renames logComplete -> checklistComplete,
lintRun -> markdownLintRun, commitCreated -> changesCommitted, jsonValid ->
validationPassed, and adds handoffPreserved + serenaMemoryUpdated.

Hardened scripts/validate_pr_review_config.py against CWE-22 path traversal
by routing the user-supplied config_path through validate_safe_path against
_PROJECT_ROOT before any open(). Matches the pattern already used in
scripts/validate_session_json.py. Added 3 CLI tests covering relative
traversal, absolute paths outside the repo, and the default-path happy path.
Removed unused pytest/Path imports flagged by ruff.

Python Security Checks (pip-audit CVE-2026-3219) is resolved by the rebase
on origin/main, which now ignores the CVE per workflow update in #1779.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* test(validate-config): add bool and null-byte/control-char coverage

Address two CodeRabbit follow-ups on PR #1671:

- Add explicit `bool` regression tests for the integer fields
  (all_open_max_prs, completion_gate_max_retries, per_pr_max_response_tokens).
  Python's `bool` is a subclass of `int`, so the validator's explicit `bool`
  exclusion needs a test to prevent silent regression.
- Add a `TestPathValidationHardening` class that exercises the new
  `validate_safe_path` integration against null-byte injection, control-
  character injection, traversal, and absolute-outside-root inputs. Null
  bytes cannot reach the CLI through subprocess argv (the OS rejects them),
  so those cases call `validate_safe_path` directly. Control chars resolve
  to a non-existent path; the CLI then exits 2 with a "not found" message,
  which is acceptable rejection behavior.

Tests: 43 passed (8 new), 0 failed.

Refs #1668

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: rjmurillo-bot <rjmurillo-bot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: Test <test@test.com>
Co-authored-by: Richard Murillo <6811113+rjmurillo@users.noreply.github.com>