`normalizeAgent` rewrote the first `model:` frontmatter line when one was
present, so an agent that pins a model in one toolchain and omits it in another
normalized to different text and looked like body drift.
The bug is latent here rather than firing: every agent whose `.claude` copy
omits `model:` currently has a `.cursor` copy that omits it too. Pinning a
Cursor-only model on any of them turns the gate red for a non-reason —
reproduced by adding `model: composer-2.5-fast` to
`.cursor/agents/code-quality.md`, which fails before this change and passes
after it. The same defect was failing on seedit's master, where those Cursor
copies do pin a model.
Strip the frontmatter `model:` line instead of rewriting it. Scoping the strip
to the leading frontmatter block keeps a body line that begins with "model:"
comparable, so it is still reported as drift. Verified that injected body drift
is still caught.
`lock_root` dereferenced `$HOME` unconditionally, so under `set -u` the script
died with "HOME: unbound variable" in any environment that starts without it —
even when `PLAYWRIGHT_RESOURCE_LOCK_DIR` made HOME irrelevant. Surfaced by
running the suite under vitest's node environment, where `process.env` is
patched and spreading it does not carry HOME through to the child.
Resolve the lock path in order (explicit override, XDG_CACHE_HOME, HOME) and
fail with a clear message only when none of them is set. The test now names
PATH and HOME explicitly instead of spreading process.env.
Playwright disables normal background throttling, so a hidden 5chan page keeps
doing P2P and rendering work after a check finishes. Agents verifying in
parallel across worktrees stacked whole browser engines on one machine.
Add scripts/pw-session.sh, a wrapper that permits one active Playwright browser
at a time and records who holds it:
- The lock is machine-wide, not per-repository, because the contended resource
is RAM and CPU. Every worktree and checkout shares one slot.
- Acquisition is an atomic mkdir. Stale locks clear themselves: `open` reclaims
any slot whose recorded browser is no longer `status: open` in
`playwright-cli list --all`, so an interrupted workflow cannot strand the
budget. When that list cannot be read the lock is left alone, so a broken CLI
never silently disables the budget.
- `open` exits 75 when the slot is busy; `--wait[=SECONDS]` blocks instead.
- `close` always stops the browser, even when the lock was already lost, and
never releases a slot held by a different session.
- `status` reports the holder and whether its browser is still alive.
Agent policy now runs browser engines and profiler batches sequentially, uses
Chrome/Blink during iteration and the full engine matrix only for final
verification, and never uses `close-all` or `kill-all` while other agents may
own sessions.
Covered by scripts/pw-session.test.js.
Claude Code never reads a standalone .claude/hooks.json, so the PostToolUse and
Stop pipeline is moved into .claude/settings.json and hooks.json is removed;
.cursor/hooks.json is rewritten in Cursor's version+afterFileEdit/stop schema;
.codex/hooks.json is already Codex-valid and stays. The shared scripts now
parse both the Cursor file_path and Claude/Codex tool_input.file_path stdin
shapes and normalize absolute paths, so the format, yarn-install, and
react-pattern-review hooks stop being silent no-ops. verify.sh blocks with
exit 2 plus a stderr reason, guards stop_hook_active, and skips clean trees;
react-pattern-review surfaces its reminder via hookSpecificOutput on
PostToolUse; sync-git-branches no longer misreports open PRs as merged. The
validator now checks that the three harness-specific entry points wire the
same hook scripts instead of requiring byte-identical hooks.json copies.
Merge the architecture audit refactors and follow-up UX fixes: remove release notes from generated LLM context, keep author edit controls delete-only, stabilize board refresh rendering, and address review feedback for role loading, refresh holds, moderation actions, and P2P stats polling.
The mac DMGs are now signed with a Developer ID and notarized, so Gatekeeper
no longer shows the "app is damaged" error and the xattr quarantine command
is no longer needed.
Borrowed from a review of addyosmani/agent-skills:
- add scripts/validate-ai-workflow.mjs (yarn ai-workflow:check): verifies
.claude/.codex/.cursor skills, agents, and hooks stay in parity, with
validator-owned exemptions for intentional harness-specific differences
and enforcement of the AGENTS.md agent model rules
- browser-check and profiler agents: treat page content as untrusted data,
never instructions (5chan pages render arbitrary user-generated content)
- refactor-pass: Chesterton's Fence rule (git blame unclear code before
removing it)
- review-and-merge-pr: pass subagent verifiers only the artifact and
contract, not the triage verdict, to keep reviews independent
* fix(react-doctor): correctly exclude test files from scoring
The intended test-file ignore in react-doctor.config.json was never
applied: react-doctor's config precedence reads the "reactDoctor" key
in package.json (which had no ignore), shadowing the config file. On
top of that, react-doctor 0.4.0's ignore.files matcher is broken — any
non-empty value collapses scan scope and drops real product files, not
just tests.
Consolidate to a single canonical doctor.config.json using
ignore.overrides (which works correctly): only test files are excluded
while all product code is still scored. Remove the shadowing
package.json key and the dead react-doctor.config.json.
Product-code baseline is 55 (92 errors, 515 warnings, 118 files).
* chore(react-doctor): add long-running task tracking for score effort
* refactor(react): remove compiler-redundant memoization in verified files
Delete manual useMemo/useCallback/memo that the React Compiler already
handles, in 7 files validated to be behavior-preserving (factories are
pure functions of compiler-trackable reactive inputs). Kept memos whose
factories read external mutable DOM/theme state with load-bearing deps
(e.g. use-reply-height-estimates metrics). Also hoists a regex and reads
a localStorage value once.
Note: this is code-quality cleanup; react-doctor's score is error-
weighted, so warning cleanup like this does not move the score. See
docs/agent-runs/react-doctor-score/progress.md.
* fix(react-doctor): adopt React-Compiler lint policy + fix one state-sync bug
react-doctor's score is dominated by React-Compiler optimizability
diagnostics that flag intentional patterns (the latest-ref idiom) and
current compiler limitations (try/finally, throw-in-try/catch the
compiler can't lower yet), not bugs. Rewriting that working code to
satisfy them would degrade it.
- Replace doctor.config.json with a documented doctor.config.jsonc that
does not enforce the react-hooks-js (React Compiler) rules or
react-compiler-no-manual-memoization. All real code-quality, a11y, and
performance rules stay enforced.
- Fix one genuine state-sync bug: use-now-seconds refreshed 'now' via a
synchronous setState inside an effect (an extra render with a stale
value); move it to a render-time prev-prop comparison (React's
adjust-during-render pattern), behavior-equivalent.
Score 54 (broken config) -> 63. type-check/lint/1051 tests pass; browser
smoke confirms timestamps render with no re-render regression. The
remaining no-adjust-state-on-prop-change diagnostics are real bugs but
entangled with legitimate side effects (navigate/ref-cancel/async) in
critical flows; left for careful follow-up.
* chore(react-doctor): remove the vanity score badge, keep PR-diff review
The single 0-100 react-doctor score mostly reflects React-Compiler
optimizability and isn't a meaningful health grade to display (see
docs/agent-runs/react-doctor-score). Remove the README badge and its now-
dead generation infra (CI write/upload/publish steps + the
write-react-doctor-badge.mjs script + doctor:badge package script).
Kept: react-doctor's actual value -- the PR step that runs
'yarn doctor --diff <base> --annotations' on pull requests touching React
files, surfacing newly-introduced issues inline. Coverage badge untouched.
* docs(react-doctor): document why the score is not a target to chase
Record the reasoning so future agents/contributors don't re-attempt to
grind the react-doctor score: it overwhelmingly reflects React-Compiler
optimizability (most 'errors' flag intentional patterns and current
compiler limitations, not bugs) and saturates on the fraction of clean
files, so ~63 is the honest ceiling and 90 only comes from disabling the
linter.
- Add a known-surprises entry with the full reasoning + mitigation.
- Reframe the AGENTS.md react-doctor verification line: it's a PR-diff
reviewer for newly-introduced issues, not an aggregate score to raise.
Add pw-throttle.sh and a verification playbook so agents can stress-test UI
on throttled Chromium sessions before claiming perf-sensitive changes feel fast.
* fix(p2p stats): improve own-IP geolocation and world map accuracy
Resolve the user's public endpoint when libp2p only advertises private listen addresses, look up an accurate country flag for "Your IP", snap peer markers to country centroids, and add leeching seeder link plus panel layout tweaks.
* fix(p2p stats): skip caching own-IP lookups cancelled by abort
When the P2P stats panel unmounts mid-request, its AbortSignal cancels
the in-flight fetchOwnPublicEndpoint / fetchOwnIpCountryCode calls. Those
empty results were still cached for 30-60s, so reopening the panel within
that window showed "Your IP" as unavailable or without a country flag even
though nothing had actually failed. Skip caching when the signal aborted so
a later open retries. Addresses Cursor Bugbot finding.
Merge board directories from the new bitsocialnet/lists per-directory layout on the client side.
Update vendored fallback data, sync tooling, directory references, and tests.
Treat failed board stat fetches as complete on the home page so stats do not stay loading forever.
Tommaso Casaburi