* ci: audit closed-issue delivery claims — cited PRs must actually be merged
Follow-up to #3223: an issue closed as 'fixed in #N' while #N was closed
without merging (or does not exist) leaves a completion claim that does not
hold. The weekly maintenance workflow now runs a read-only audit over issues
closed in the last 30 days, extracts delivery-claim PR references from issue
bodies and close-window comments, verifies each cited PR's merge state, and
reports findings to the step summary.
Design constraints:
- Read-only: no comments, no closes, no labels; findings are advisory and
the run always stays green so the weekly audit never blocks maintenance.
- Unverifiable counts as unmerged: a 404 PR or an API failure fails the
verdict, because an unverifiable claim is exactly the failure mode.
- Pure decision functions are exported for tests; a first live run against
this repository surfaced one real finding (#3408 cites #3060, which does
not resolve), included here as the motivating sample.
8 unit/contract tests: extraction, windowing (close comments included),
the #3223 unmerged-citation failure mode, the all-clear summary, workflow
contract, and a no-write guard.
* ci: rework audit evidence contract — claims vs context, four PR states, visible completeness
Revision per the #3522 review (reviewer ran the CLI through 19 controlled
cases plus a bounded live check of #3408):
- Bind findings to asserted delivery claims. Closing wording ('Delivered
through #N'), claim verbs (fixed/shipped/landed/resolved/...), and bare
'PR #N' inside closing comments are claims; historical mentions, quoted
lines, fenced examples, negated claims, and tracking-only references are
reported as contextual references and never produce findings. On the live
#3408 the audit now recognizes the closing claim #3500 (merged) instead of
flagging the historical #3060 text.
- Separate observed states: merged / open-unmerged / closed-unmerged /
unknown (404, rate limit, transient). Unknown claims are advisory manual-
verification items, not failed claims; each row carries the claim snippet,
its source link, and the observation time. Checked claims are listed even
when passing, so the report is traceable rather than an opaque all-clear.
- Paginate issue and comment collection (--paginate --slurp), bound comment
requests (200) and PR checks (100), and surface skipped pages, rate limits,
and retrieval failures as an explicit incomplete audit instead of an empty
all-clear. A live-run maxBuffer overflow on the paginated issue list was
caught and fixed (64MB); the audit reported it as incomplete before the fix.
- Scope the job to read access: job-level contents/issues/pull-requests read.
13 contract/unit tests including a CLI fixture (stub gh) that places the
relevant issue on page two and forces a comment-fetch failure.
* ci: report incomplete audit at PR check cap
---------
Co-authored-by: chenhz01 <chenhz01@users.noreply.github.com>
Co-authored-by: haelyra <49814733+haelyra@users.noreply.github.com>
Bash removes a backslash-newline before it reads the line, so the character before the pair
decides whether a following # starts a word. `echo \` then `# <<EOF` on the next line is a
comment, and the command after it runs; `echo foo\` then `#bar` stays one word. Count the
backslashes before a blank instead of remembering the last escape, which also took the
continuation for an escaped character.
A carriage return is part of a bash word and line, so after an LF delimiter a line ending in
\r no longer ends the body early and exposes the body as commands. In an expanded body, a line
ending in an escaping backslash is joined with the next before it is compared, so `EO\` and
`F` end an unquoted <<EOF and the command after it is checked. A # after an escaped blank, as
in `echo foo\ # && npm run dev`, is part of the word, not a comment.
Invalid UTF-8 in a $'...' delimiter became U+FFFD, which a body line could match although Bash compares raw bytes. Such a delimiter now matches no line, so its body runs to the end of the command, as in Bash. The bytes were also spread into one push() call, which threw for a delimiter of 200,000 characters. The hook then failed open. The bytes are now collected as buffers and concatenated once.
Bash decodes $'...' quoting, so <<$'E\x4fF' ends at an EOF line. The splitter kept the escape, ended the body at a literal E\x4fF line, and exposed the rest of the body as commands. The delimiter word is now read by readWord(), which decodes $'...' as Bash does, \' included.
Bash removes a backslash before a newline, but not one before a carriage return. Removing backslash-CRLF joined two lines of the message, and their text was then read as the subject.
In double quotes, Bash drops the backslash before $, `, ", \ and a newline, so git receives fix: say "hello" for -m "fix: say \"hello\"". The check kept the backslashes, so an escaped backslash before a newline counted twice. These escapes are now decoded, and a backslash before any other character stays. The test that pinned the escaped quotes now expects the message git records.
Bash reads <<$'EOF' as the delimiter EOF, but the splitter took the $ into the word and waited for a $EOF line, so the body ran on and swallowed the commands after it. A $ before a quote now only starts the quoting.
Bash removes a backslash-newline inside double quotes, so git never sees it, but the check kept both characters. A message that starts with one had a subject of just the backslash and got a false format warning. Line continuations are now dropped from a double-quoted message before its subject is read. An escaped backslash before a newline stays, and a here-document body is read as written.
The quoted-message pattern took an escape as a backslash and one character other than a newline. A here-document body with a line ending in a backslash, such as a Windows path, therefore matched no -m at all, and the message was not checked. A backslash may now escape a newline too.
Bash expands a here-document body unless part of its delimiter is quoted, and runs the $(...) and backquoted substitutions in it, with quotes as plain text. Leaving the whole body out of the segments hid those commands, so npm --version <<EOF with '$(npm run dev)' in the body was allowed. The commands of these substitutions are now segments of their own. An escaped substitution, or a body whose delimiter is quoted (<<'EOF', <<\EOF), stays data.
The closing pattern accepted spaces and tabs around the delimiter for both forms, so a body line such as ' EOF' ended the message early. Bash ends <<EOF only at a line that is exactly EOF, and <<-EOF also after leading tabs. The body is now read line by line from the command, and the existing message pattern is back to what it was on main.
A body attached to the segment of its command was still searched by the dev-server check. For npm --version <<EOF && cat, the body line npm run dev made the npm segment look like a dev server. A body is data, so it is no longer part of any segment. An empty delimiter (<<'') is a valid opener again, and its body ends at the first empty line.
The commit-message checks read the whole -m value. A multi-line message was measured against the 72-character subject limit as a whole, and a body ending in a period got the trailing-period warning. A message written as -m "$(cat <<'EOF' ... EOF)", the form Claude Code uses, was checked as the literal $(cat <<'EOF' text, so every such commit got the format and length warnings. The here-document body is now read as the message, and the format, length, capitalization and period checks apply to its subject line, the first line git keeps.
The opener pattern read only an identifier, so <<EOF/END stored EOF, <<E"OF" stored E and <<123 was no opener at all. In each case the body was split into segments, or the line after it was swallowed. The delimiter is now read as a shell word with its quotes and backslashes removed, as Bash does. A body that follows an operator at the end of its line (cat <<EOF &&) was split off as a segment of its own. It now goes with the command that opened it.
A << in a comment (echo ok # <<EOF) or an identifier shift ($((x << n))) was read as a here-document opener, so the following lines, npm run dev included, were swallowed into its body. A comment now runs to the end of its line and an arithmetic expression is copied whole, so neither opens a here-document nor splits at an operator. A delimiter line ending in \r (CRLF) now ends the body too.
splitShellSegments() split at &&, ||, ; and & but not at a newline, so a multi-line command such as "cd app\nnpm run dev" was one segment led by cd, and the dev-server block let the dev server through. A newline now ends a segment. The body of a here-document (<<EOF, <<-EOF, <<'EOF') stays in the segment of the command that reads it, so a README written with cat <<EOF that mentions npm run dev is not blocked; a here-string (<<<) or a shift such as $((x << 2)) is not read as one.
Prettier's config searcher also reads .prettierrc.json5, .prettierrc.ts/.mts/.cts and prettier.config.ts/.mts/.cts. resolve-formatter knew none of them, so a project configured only through prettier.config.ts was never formatted by post-edit-format, quality-gate or the Stop batch. config-protection's rc pattern stopped at .yaml/.toml, so an existing .prettierrc.ts or .prettierrc.json5 could be edited freely.
The Stop typecheck reported every tsc line that contained the edited path anywhere. A root index.ts has the relative path `index.ts`, which every `src/index.ts(...)` diagnostic contains, so a clean root file was reported with the errors of src/index.ts. The same happened to src/utils.ts with web/src/utils.ts, and to any file named in another file's message.
Compare the path that starts the diagnostic (`path(line,col): `), keeping the slash normalization for Windows.
With diff.relative=true, `git diff --name-only` reports paths relative to the working directory, so joining them to the repository root pointed at files that do not exist, and a session in a subdirectory again got no console.log warning. getGitModifiedFiles now runs `git -c diff.relative=false diff --name-only HEAD`. The -c form also works on Git versions older than --no-relative.
readToken dropped every backslash inside double quotes, while bash drops it only before $, `, ", \\ or a newline. `"echo ready\;"` therefore reached the tmux check as `echo ready;`, read as a command separator, and `tmux send-keys -t dev "echo ready\;" "npm run dev" Enter` was blocked although it types into a pane.
check-console-log took the paths from `git diff --name-only HEAD`, which are relative to the repository root, and checked them with fs.existsSync from the hook's working directory. In a session started in a subdirectory, such as a monorepo package, no path existed from there, so the hook never warned. Resolve each path from the repository root. The exclusion patterns and the warning still use the root-relative path.
tmux also separates commands at a `;` that ends an argument, so `new-session -d 'npm run dev;' run-shell 'npm run dev'` runs the second dev server without a pane. Split there as well. An escaped `\;` at the end of an argument stays literal, as in tmux.
Two forms got the pane exemption although they run the dev server without a pane.
`tmux -vL new-session run-shell ...`: in a cluster, the last option that takes a value (f, L, S, T) takes the next token, here the socket name `new-session`. A `c` anywhere in a cluster (`-c'npm run dev'`) runs a shell command without a pane.
`tmux new-session -d \; run-shell "npm run dev"`: only the first command of a sequence was checked. The sequence is now split at `;`, and it is exempt only when every command in it that runs the dev server is a pane command. This also closes the same hole for the plain `tmux new-session ... \; run-shell`, which the old regex exempted.
The block allowed a dev server only when the segment began with `tmux new`, `new-session`, `new-window` or `split-window`. It therefore blocked dev servers that do start in a tmux pane, the way its own message asks: `tmux -L work new-session ...` and `tmux -f conf new-session ...` (tmux options before the command), `/opt/homebrew/bin/tmux new-session ...` and `env TERM=... tmux new-session ...`, the aliases `neww` and `splitw`, and `tmux send-keys -t dev 'npm run dev' Enter` or `respawn-pane` for a session that exists.
Read the tmux command from the parsed segment instead, after the leading command word and tmux's own options, and allow the commands that run their shell command in a pane. `tmux run-shell` and `tmux -c` run it without one and stay blocked.