Skip to content

Explain argument-rejected run-tests receipts and guard worktree links (#2036) - #2048

Merged
wwind123 merged 1 commit into
mainfrom
agent-loop/managed-2036
Nov 20, 2026
Merged

wwind123 merged 1 commit into
mainfrom
agent-loop/managed-2036

Conversation

@wwind123

Copy link
Copy Markdown
Owner

Summary

Implements the approved plan for #2036.

Root cause. _is_playwright_test_runner only accepts long options, and accepts value options only in --name=value form. The coder's -g '<pattern>' (a short alias with a separate value) failed that check in every variant: npx, bare playwright, and npm run <script> -- ... -g ..., whose appended args go through the same recognizer. The shared node_modules symlink had nothing to do with it. The broker persists only the coarse unrecognized-launcher reason, so the client printed "unrecognized test launcher" for a launcher family the prompt lists as citable.

Changes

  • Specific receipt reason (test_runtime.py, cli.py):
    • New static, side-effect-free explain_uncitable_launcher.
    • _is_playwright_test_runner and _is_node_test_runner now delegate to shared _first_rejected_*_argument walks. Their accepted sets are unchanged, so the explanation can never disagree with the recognizer.
    • The run-tests notice names the rejected argument and the citable spelling, for example -g pat -> --grep=pat.
    • Other causes it names:
      • code-loading --config, which is never citable, including a scratch config outside the checkout
      • zero-test or interactive options
      • the -- separator
      • unknown options
      • a non-empty NODE_OPTIONS/PW_TEST_REPORTER
      • a missing local Playwright binary for npx playwright
      • a package script with a pre/post hook or shell syntax
    • An argument appended via npm run <script> -- ... is called out as appended, with the advice to cite the unfiltered script run.
    • Unknown or unresolvable commands keep the generic text.
  • Wording: CITABLE_TEST_RUNNERS_SUMMARY now states the allow-list / --name=value rule, and the post-auth SUITE_START_UNRECOGNIZED_DIAGNOSTIC no longer says the launcher is unrecognized. No persisted value changed.
  • Coder link guidance (prompt_support.py):
    • Implementation prompts name the coder checkout's own registered --worktree-link paths as shared operator infrastructure: do not replace, move, copy over or delete them; report dependency problems instead.
    • The names come from the checkout's registration, never from config.worktree_links alone.
    • Under --worktree-setup-pr-local the existing note stays, with its gating unchanged, and now also forbids moving and copying over. It is the only paragraph naming those links.
  • Pre-launch link check (checkout_verification.py, local_test_evidence.py):
    • Before reserving or launching anything, the broker refuses a request whose checkout has a registered link (or private environment) that is no longer intact. The refusal is a no-test cancelled frame that names the link.
    • The check is fail-open, never poisons the checkout, and leaves the guard unchanged.
    • The snapshot error text now includes the link name.
  • Docs: docs/local_agent_loop.md, docs/architecture/execution-and-concurrency.md, and the matching entry in its decisions log. Two sentences in the architecture doc that duplicated the CLI guide were dropped to stay under the 64 KiB doc cap.
  • Fixture hashes and golden prompts: updated for the intentionally changed bodies and summary text.

Tests

  • tests/test_test_runtime.py:
    • explainer pins for the bare, local-bin, npx and npm run -- ... forms, plus each rejection category
    • parity against the existing playwright, node and npx fixture tables, which are now shared module constants
    • None for unknown or unresolvable commands
  • tests/test_agent_loop.py: the CLI notice names -g/--grep=pat and never says "unrecognized test launcher", with the persisted reason and with no reason (older broker). A malformed package.json keeps the generic text.
  • tests/test_prompts.py:
    • registered links are named for the coder only, never for reviewers
    • no guidance for an unregistered checkout or in mixed mode
    • pr-local before and after a rebuild: one paragraph only, with the new move/copy-over wording
  • tests/test_local_test_evidence.py:
    • a link replaced by a directory, retargeted, or removed is refused with the named message: nothing is launched, no evidence row is recorded, nothing is reserved, and the checkout is not poisoned
    • an intact link runs normally
    • the pre-check fails open on lookup errors
    • the CLI prints the refusal and never falls back to a local run
  • Full suite: agent-loop run-tests -- <venv python> -m pytest tests/ -q

Fixes #2036

AGENT_MANAGED_CI_UNPROTECTED_OVERRIDE_V1 nonce=OoejcBVFLEA-maK_SVxy0Uf8

🤖 Generated with Claude Code

-- Anthropic Claude

…#2036)

- A static explainer shares the recognizer's playwright/node argument walk
  and names the rejected argument (for example `-g pat` -> `--grep=pat`,
  code-loading `--config`, appended `npm run <script> -- ...` args) instead
  of calling a citable launcher family "unrecognized". The allow-list and
  persisted suite_start reasons are unchanged; the notice is advisory only.
- The citable-runner summary and post-auth diagnostic describe the
  `--name=value` allow-list rule.
- Coder prompts name the coder checkout's registered --worktree-link paths
  as shared operator infrastructure not to be replaced, moved, copied over
  or deleted; the pr-local note also forbids moving/copying over.
- The test broker refuses a request before reservation or launch when a
  registered link or private environment is not intact, naming it. The
  pre-check is fail-open and never poisons the checkout.

Fixes #2036

Agent-Issue-Provenance: v1 repo=wwind123/coding-review-agent-loop issue=2036 flow=approved plan=e71f1299f12ce9e5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@wwind123 wwind123 added the agent-loop-managed agent-loop managed CI label Nov 20, 2026
@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI authorization record (machine-readable). Binds issue #2036 to pull request #2048 at head 30bfe51. Not an agent response; keep this comment.

@wwind123

Copy link
Copy Markdown
Owner Author

Expected closing issues for PR #2048: #2036.

Origin flow: approved-plan-implementation
Contract hash: be23edc842d0f32704eb714e34c064dabff7675437ce716c67cd346921c164ad

-- coding-review-agent-loop

@wwind123

Copy link
Copy Markdown
Owner Author

Issue implementation

Implemented the approved #2036 plan in draft PR #2048. Root cause: the playwright allow-list accepts value options only in --name=value form, so -g pat was rejected (including via npm run appended args), and the client reported the coarse persisted reason as 'unrecognized test launcher'. Changes: (1) a static explainer that shares the recognizer's argument walk and names the rejected argument and its citable spelling, with updated summary and post-auth diagnostic wording; the allow-list and persisted reasons are unchanged. (2) Coder prompts name the checkout's registered --worktree-link paths as off limits; the pr-local note now also forbids move/copy-over. (3) The broker refuses a request before launch, naming the replaced link; the check is fail-open and never poisons. Test rationale: the focused run covers each changed module (test_test_runtime for the explainer, test_agent_loop for the CLI notice, test_local_test_evidence for the broker pre-check, test_prompts for the prompt guidance and goldens, and the docs/architecture checks for doc edits). The full tests/ run was done because the plan's one-shot acceptance criteria explicitly require it, and because shared fixtures (move-manifest hashes, prompt goldens) changed; it passed (25999 passed). An architecture-doc sentence pair duplicating the CLI guide was dropped to stay under the 64 KiB doc cap.

Result

Pull request reported: #2048.

Tests run

  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -n 0 tests/test_test_runtime.py -k '2036 or playwright or node' (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_protocol.py tests/test_protocol_risk_move.py tests/test_module_split_imports.py tests/test_worktree_pr_env.py tests/test_source_file_size.py (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -n 0 tests/test_prompts.py -k '2036 or root_only_under_limit or pr_local_environment_note or citable' (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -n 0 tests/test_docs_guidance.py -k 'share_one_shape or default_render_injects' (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q (agent-loop instrumented; whole-command timeout 1800s)

Test observation receipts

  • '<path-sha256:d094488e6a9166dc>' -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py — receipt 66a88962c8464a22b6635e0c0d9205d7 — current-result (verified against the parent journal; CI remains authoritative)
  • '<path-sha256:d094488e6a9166dc>' -m pytest tests/ -q — receipt b1ee951409f2491da33b3b0467b6eee1 — current-result (verified against the parent journal; CI remains authoritative)
  • '<path-sha256:d094488e6a9166dc>' -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_protocol.py tests/test_protocol_risk_move.py tests/test_module_split_imports.py tests/test_worktree_pr_env.py tests/test_source_file_size.py — receipt 6867cefbc2ba435f81f418e034e5fd5a — historical failure at an obsolete tree (not an outstanding obligation) failed
  • '<path-sha256:d094488e6a9166dc>' -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py — receipt b9b0bbf9173245edba39444d6c519946 — historical failure at an obsolete tree (not an outstanding obligation) failed
  • '<path-sha256:d094488e6a9166dc>' -m pytest -q -n 0 tests/test_docs_guidance.py -k '<arg-sha256:162a3b3dfb48616f>' — receipt 4a184890876c4b339c172dab7aa236e3 — historical failure at an obsolete tree (not an outstanding obligation) failed

Risk-based mode and transition test matrix evidence

  • Matrix identity: 63844a8750650576c56e55098a914d086210bfd19fb2d2d0721b2add35a6d392
Full matrix evidence (6 rows)
  • legacy-broker-no-reason — verified
    • Tests: tests/test_agent_loop.py::test_run_tests_notice_names_the_rejected_playwright_argument_2036, tests/test_agent_loop.py::test_run_tests_notice_keeps_the_generic_text_for_a_malformed_package_json_2036
    • Locations: tests/test_agent_loop.py
    • Workflow path: run-tests CLI with a broker frame that has suite_start=unknown and no suite_start_reason (older-broker parametrization)
    • Expected outcome assertions: The static fallback prints the same specific -g/--grep=pat notice; A malformed package.json keeps the generic unrecognized notice
    • Forbidden-effect assertions: No crash on a malformed package.json; exit code 0
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]
  • link-replaced-refusal — verified
    • Tests: tests/test_local_test_evidence.py::test_broker_refuses_a_replaced_worktree_link_before_launch_2036, tests/test_local_test_evidence.py::test_broker_runs_normally_with_an_intact_worktree_link_2036, tests/test_local_test_evidence.py::test_cli_prints_the_replaced_link_refusal_and_never_runs_locally_2036, tests/test_local_test_evidence.py::test_replaced_worktree_link_precheck_is_fail_open_and_names_private_paths_2036
    • Locations: tests/test_local_test_evidence.py
    • Workflow path: real BrokerServer over its socket in a registered run worktree whose node_modules link is replaced by a directory, retargeted or removed; plus the run-tests CLI against that broker
    • Expected outcome assertions: The cancelled no-test frame diagnostic names node_modules as shared operator infrastructure; The CLI prints 'no test ran (cancelled)' with the named refusal; An intact link runs and passes
    • Forbidden-effect assertions: The inner process never runs (marker absent), the journal is empty, nothing is reserved; has_poisoned_entry stays False and there is no local fallback run
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]
  • npm-appended-args-notice — verified
    • Tests: tests/test_test_runtime.py::test_explainer_names_the_rejected_argument_appended_to_a_package_script_2036, tests/test_test_runtime.py::test_explainer_names_a_rejected_package_script_body_2036
    • Locations: tests/test_test_runtime.py
    • Workflow path: static explainer over package-script resolution: package.json body playwright test, argv npm run s -- f.spec -g pat
    • Expected outcome assertions: The reason names -g in an argument appended to npm run s and suggests --grep=pat; It says to cite the unfiltered npm run s run and report this run in tests_run; The unfiltered npm run s is still recognized
    • Forbidden-effect assertions: The reason does not contain 'unrecognized'
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]
  • prompt-links-present-absent — verified
    • Tests: tests/test_prompts.py::test_coder_prompt_names_registered_worktree_links_as_off_limits_2036, tests/test_prompts.py::test_coder_prompt_has_no_link_guidance_for_an_unregistered_checkout_2036, tests/test_prompts.py::test_pr_local_coder_prompt_names_each_link_once_without_the_shared_paragraph_2036, tests/test_prompts.py::test_pr_local_environment_note_only_with_the_flag_and_after_a_rebuild
    • Locations: tests/test_prompts.py
    • Workflow path: coder workdir guidance and implementation prompt rendering with registered, unregistered (no links and mixed mode) and pr-local pre-/post-rebuild registrations
    • Expected outcome assertions: A registered node_modules is named with do-not-replace/move/copy-over/delete text in coder prompts; Under pr-local, before and after a rebuild, the pr-local paragraph is present with the move/copy-over wording and is the only paragraph naming the link
    • Forbidden-effect assertions: No guidance for an unregistered checkout even when config.worktree_links is set; Reviewer prompts carry no guidance, and pr-local paths are never called shared operator infrastructure
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]
    • Caveats: The pre-existing pr-local paragraph itself names the link twice, so the test pins that the full guidance names it exactly as often as that paragraph alone, with no second paragraph.
  • pw-short-alias-notice — verified
    • Tests: tests/test_agent_loop.py::test_run_tests_notice_names_the_rejected_playwright_argument_2036
    • Locations: tests/test_agent_loop.py
    • Workflow path: run-tests CLI with a broker result frame carrying suite_start=unknown and the persisted unrecognized-launcher reason, for the npx, bare local-bin and npm-appended playwright forms
    • Expected outcome assertions: Stderr names -g as a short alias and suggests --grep=pat; The notice says suite_start=unknown and that the run may be reported in tests_run
    • Forbidden-effect assertions: 'unrecognized test launcher' does not appear; Exit code stays 0 and only the timing row is recorded (no launcher health), so citability is unchanged
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]
  • recognizer-parity — verified
    • Tests: tests/test_test_runtime.py::test_explainer_and_recognizer_agree_within_supported_families_2036, tests/test_test_runtime.py::test_explainer_returns_none_for_unknown_or_unresolvable_commands_2036, tests/test_test_runtime.py::test_playwright_test_is_a_recognized_inner_launcher, tests/test_test_runtime.py::test_playwright_unsafe_forms_stay_unrecognized, tests/test_test_runtime.py::test_node_test_runner_is_a_recognized_inner_launcher, tests/test_test_runtime.py::test_node_without_builtin_test_runner_stays_unrecognized, tests/test_test_runtime.py::test_npx_playwright_unsafe_forms_stay_unrecognized
    • Locations: tests/test_test_runtime.py
    • Workflow path: static recognizer versus explainer over the shared playwright/node/npx fixture tables, the new rejection cases and the =-form filtered commands, with resolvable context
    • Expected outcome assertions: The explainer returns None for every recognizer-accepted argv, including --grep=pat --repeat-each=200; Every supported-family rejection gets a non-None reason; make test, npx --test, npx jest, a missing script and a malformed package.json get None
    • Forbidden-effect assertions: The existing accept/reject fixture tables pass unchanged, so the allow-list is not relaxed; No accepted command gets a reason, and no unknown head gets a family-specific reason
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q tests/test_test_runtime.py tests/test_agent_loop.py tests/test_local_test_evidence.py tests/test_prompts.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [66a88962c8464a22b6635e0c0d9205d7; current-result]; /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q [b1ee951409f2491da33b3b0467b6eee1; current-result]

Risk-matrix coverage map

Orchestrator completeness check (existence and level only; whether each cited test really exercises its row is for the reviewer to judge).

  • pw-short-alias-notice: tests tests/test_agent_loop.py::test_run_tests_notice_names_the_rejected_playwright_argument_2036, tests/test_agent_loop.py; declared level unit vs required unit
  • npm-appended-args-notice: tests tests/test_test_runtime.py::test_explainer_names_the_rejected_argument_appended_to_a_package_script_2036, tests/test_test_runtime.py::test_explainer_names_a_rejected_package_script_body_2036, tests/test_test_runtime.py; declared level unit vs required unit
  • legacy-broker-no-reason: tests tests/test_agent_loop.py::test_run_tests_notice_names_the_rejected_playwright_argument_2036, tests/test_agent_loop.py::test_run_tests_notice_keeps_the_generic_text_for_a_malformed_package_json_2036, tests/test_agent_loop.py; declared level unit vs required unit
  • recognizer-parity: tests tests/test_test_runtime.py::test_explainer_and_recognizer_agree_within_supported_families_2036, tests/test_test_runtime.py::test_explainer_returns_none_for_unknown_or_unresolvable_commands_2036, tests/test_test_runtime.py::test_playwright_test_is_a_recognized_inner_launcher, tests/test_test_runtime.py::test_playwright_unsafe_forms_stay_unrecognized, tests/test_test_runtime.py::test_node_test_runner_is_a_recognized_inner_launcher, tests/test_test_runtime.py::test_node_without_builtin_test_runner_stays_unrecognized, tests/test_test_runtime.py::test_npx_playwright_unsafe_forms_stay_unrecognized, tests/test_test_runtime.py; declared level unit vs required unit
  • prompt-links-present-absent: tests tests/test_prompts.py::test_coder_prompt_names_registered_worktree_links_as_off_limits_2036, tests/test_prompts.py::test_coder_prompt_has_no_link_guidance_for_an_unregistered_checkout_2036, tests/test_prompts.py::test_pr_local_coder_prompt_names_each_link_once_without_the_shared_paragraph_2036, tests/test_prompts.py::test_pr_local_environment_note_only_with_the_flag_and_after_a_rebuild, tests/test_prompts.py; declared level unit vs required unit
  • link-replaced-refusal: tests tests/test_local_test_evidence.py::test_broker_refuses_a_replaced_worktree_link_before_launch_2036, tests/test_local_test_evidence.py::test_broker_runs_normally_with_an_intact_worktree_link_2036, tests/test_local_test_evidence.py::test_cli_prints_the_replaced_link_refusal_and_never_runs_locally_2036, tests/test_local_test_evidence.py::test_replaced_worktree_link_precheck_is_fail_open_and_names_private_paths_2036, tests/test_local_test_evidence.py; declared level integration vs required integration
  • Status: complete at 30bfe51 (deterministic completeness check passed)

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI resume provenance audit record (machine-readable). Records resume provenance for managed CI on pull request #2048 at head 30bfe51. Not an agent response; keep this comment.

AGENT_MANAGED_CI_UNPROTECTED_OVERRIDE_V1 nonce=JOWrX2REVJXrwPAv4bYmuOayw0T5NQAJ repo=wwind123/coding-review-agent-loop base=main head=30bfe5145a128a487cf663721b9311813e7b49c8 protection=voluntary active_label_event_id=36796365673 resume_from=7046526652 provenance_head=30bfe5145a128a487cf663721b9311813e7b49c8 generation=xXLo1TW62KGb4HL3

Resume provenance only: the prior issue-created audit is not an authorization token.

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Codex; paused Claude, Antigravity; reason: strict pre-panel fallback: architecture identity changed; scheduler metadata recovery: latest scheduler metadata is missing or invalid; scheduler metadata recovery: current-head records lack scheduler metadata; scheduler metadata is missing or invalid; primary re-invoked with full context; phase: primary; head: 30bfe51; primary: Codex; active owners: (none); force-full: False (source: none); scheduler-policy calls avoided cumulatively: 2.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Reviewed the complete base-to-head diff at 30bfe51 and verified coverage for all six risk-matrix rows. No merge-blocking defects found. CI is passing and supplied current-head focused and full-suite receipts passed. Tests were not executed during this read-only review.

-- OpenAI Codex: gpt-6.1-sol (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 1 reconciliation: settled reviewers: Codex. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 2. Phase: primary; force-full: False (source: none).

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Claude, Antigravity; paused Codex; reason: independent secondary audit after exact-head primary approval; phase: secondary-audit; head: 30bfe51; primary: Codex; active owners: (none); force-full: False (source: none); scheduler-policy calls avoided cumulatively: 2.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Independent secondary review of the full base-to-head diff at 30bfe51. (1) Receipt reason: _is_playwright_test_runner and _is_node_test_runner are now thin wrappers over the _first_rejected_*_argument walks. Each return False of the old functions maps to a non-None rejection and each return True maps to None, so the accepted set is unchanged. explain_uncitable_launcher mirrors the env-prefix, npx-rewrite and package-script resolution. It returns None for accepted commands and for unknown or unresolvable heads, and is wrapped in try/except, so it stays advisory. The CLI uses it both for the persisted unrecognized-launcher reason and for the older-broker static fallback, and the new text never says 'unrecognized test launcher'. -g pat is reported as a short alias with the --grep=pat spelling. Appended npm args are reported as appended, with the advice to cite the unfiltered script. --config/-c is reported as code-loading and never citable. (2) Link prompt: _worktree_link_guidance takes names only from the coder checkout's own registration (registered_links(workdir)) and drops names that the pr-local paragraph already covers. The pr-local paragraph keeps its gating and only gains the move/copy-over wording. Reviewer prompts are untouched. (3) Pre-check: the broker calls replaced_worktree_link(self.root) after request validation and rerun resolution, but before the journal/reservation section. On failure it sends the existing legacy no-test cancelled refusal frame, so no new outcome literal is needed. The check is fail-open on OSError/AgentLoopError and never poisons the checkout. The tests cover directory, retarget and removal, an intact link, fail-open behaviour, and the CLI path with no local fallback. Docs (CLI guide plus the architecture detail doc and its decision log) are updated. Every risk-matrix row maps to a test that exercises its path. GitHub checks are passing, and the coder's current-head full-suite receipt passed.

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Approved. The implementation addresses issue #2036 by introducing a static, side-effect-free explainer for argument-rejected test runs (e.g. Playwright short aliases, detached values, code-loading options, and appended script arguments) without altering the recognizer allow-list or persisted suite start reasons. Coder prompts clearly name registered worktree links as shared operator infrastructure not to be tampered with, while avoiding duplicate guidance under pr-local mode. The broker pre-check refuses requests with replaced worktree links fail-open before reservation or launch without poisoning the checkout. All enforceable risk matrix rows and acceptance criteria are covered by targeted tests, and doc updates maintain consistency.

-- Google Antigravity: Gemini 3.8 Flash (High)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 2 reconciliation: settled reviewers: Antigravity, Claude. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 2. Phase: secondary-audit; force-full: False (source: none).

@wwind123

wwind123 commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner Author

Managed CI authorization for exact head 30bfe51.

@wwind123
wwind123 marked this pull request as ready for review November 20, 2026 18:04
@wwind123
wwind123 merged commit 7a1ea9a into main Nov 20, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Coder replaced the shared node_modules link after a misleading 'unrecognized test launcher' receipt for npx playwright test

1 participant