ci: fail the PR when its branch or its Closes line breaks the convention - #69
Conversation
Six downstream PRs were opened with a bare "Closes #" and a branch named chore/-slug — the PR template's comment dropped, the branch's issue slot empty — and nothing rejected them. scripts/pr-lint.js now checks the head branch is <type>/<issue#>-<slug>, the body links an issue with a GitHub closing keyword (HTML comments stripped first), and the branch's number is among the linked ones; release-please and Dependabot branches are exempt. ci.yml runs it as a step-level-if github-script step on pull_request events, so the required check fails with the fix named (ADR-0008 shape). scripts/check-node-tests.sh replaces check-issue-labeler.sh and runs every scripts/*.test.js. Closes #56 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4b9b44275a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (EXEMPT_BRANCH_PREFIXES.some((p) => ref.startsWith(p))) { | ||
| return { exempt: true, problems: [], branchIssue: null, linked, cross }; |
There was a problem hiding this comment.
On this public repository, a fork author controls pull_request.head.ref; skills/release-management/SKILL.md:81 explicitly notes that a fork can reuse the predictable release-please branch name. A human PR whose branch starts with release-please-- or dependabot/ therefore takes this early return and bypasses both the required branch-format and Closes checks. Pass the PR author or actor into lint() and require the expected bot identity in addition to the prefix.
AGENTS.md reference: AGENTS.md:L48-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Agreed — same point I raised on #41. Fixed in a0161e1: exemption is by pull_request.user.login (dependabot[bot], github-actions[bot]) or by release-please's autorelease: pending label (covers the PAT-driven variant, and a fork author can apply neither); the branch prefix is no longer consulted. Test: a human on release-please--branches--main without the label gets the full lint (2 problems). Verified against REST payloads of #68 (bot author → exempt) and #11 (Dependabot → exempt).
…ch name Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a0161e1b20
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| # jobs" replays the ORIGINAL payload and would judge the old body. Never | ||
| # narrow this list: a `types:` without the three defaults is the | ||
| # "check never created" trap described above. | ||
| types: [opened, synchronize, reopened, edited, ready_for_review] |
There was a problem hiding this comment.
Blocking: When release-please uses the PAT alternative documented in skills/release-management/SKILL.md:83, the initial PR is authored by the PAT owner and autorelease: pending can only be attached after the PR has been created. The opened payload therefore lacks both exemptions and fails lint, while the subsequent labeled event does not run this workflow, leaving the first release PR red until an unrelated edit or synchronization occurs. Include labeled so the check reruns after the exemption becomes available.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in b3fc3f8: labeled added to on.pull_request.types (comment explains why); the header's superset rule still holds.
| `with the number. Any GitHub closing keyword works: Closes / Fixes / Resolves, optional colon, #N or a full issue URL${tail}`, | ||
| ); | ||
| } | ||
| if (branchIssue !== null && linked.length > 0 && !linked.includes(branchIssue)) { |
There was a problem hiding this comment.
Blocking: When a body contains both Closes #10 and Closes #11, this predicate passes as long as the branch names either issue; the added test explicitly locks in that behavior. That contradicts the one-issue-per-PR contract in skills/pr-authoring/SKILL.md:41 and permits a single PR to auto-close additional issues outside its branch linkage. Require exactly one local closing reference rather than merely checking that the branch issue is among them.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in b3fc3f8: more than one local closing reference is now a failure pointing at pr-authoring rule 5 (the same issue referenced twice is fine); the test that had locked in the opposite is inverted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b3fc3f8704
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| # AFTER opening the PR; with a PAT-authored release PR that label is what | ||
| # exempts it. Never narrow this list: a `types:` without the three | ||
| # defaults is the "check never created" trap described above. | ||
| types: [opened, synchronize, reopened, edited, ready_for_review, labeled] |
There was a problem hiding this comment.
When autorelease: pending is removed from a human-authored PR whose branch or body is malformed, no workflow fires because this list includes labeled but not unlabeled; the successful ci result created when the label was added remains attached to the same head SHA, so the now-nonexempt PR stays green. Include unlabeled so removing the exemption reruns the required check.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
| const local = !qualifier || (repo && qualifier.toLowerCase() === String(repo).toLowerCase()); | ||
| const key = `${local ? '' : qualifier + '#'}${number}`; | ||
| if (!seen.has(key)) seen.set(key, { repo: local ? null : qualifier, number }); |
There was a problem hiding this comment.
When the branch and body contain the same nonexistent number, such as fix/999999-x and Closes #999999, this code unconditionally records the syntactic reference as a local issue and lint() returns no problems; a pull-request number is accepted the same way. The new gate therefore still allows work with no backing issue and nothing for GitHub to close, so the sole local reference should be validated through the GitHub API and rejected when it is missing or represents a pull request.
AGENTS.md reference: AGENTS.md:L9-L9
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in d7fd294: run() now takes github and makes one read — issues.get on the single linked number: 404 → 'does not exist', a pull_request field → 'is a pull request, not an issue', any other error → the job fails (a check that cannot check must not pass). REST shapes confirmed live (#69 reads back with pull_request; #999999 is 404). Header wording updated: the token was already in the step's environment, so passing the client changes nothing about exposure.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d7fd294192
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| // reference qualified with this repository's own name counts as local. | ||
| function linkedIssues(body, repo) { | ||
| const seen = new Map(); | ||
| for (const m of stripHtmlComments(body).matchAll(CLOSES_RE)) { |
There was a problem hiding this comment.
When the only Closes #42 text appears inside an inline or fenced code block, matchAll still treats it as the required link and the issue lookup lets the check pass, although GitHub renders that text as code rather than a closing reference and therefore will not close the issue on merge. Parse the Markdown or remove code spans and blocks before matching closing keywords.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 31c13d1: fenced blocks (``` or ~~~) and inline code spans are stripped alongside HTML comments before matching; tests cover a fence-only body, a fence followed by prose, and inline spans around a real reference. 65k-char adversarial body strips in 0.13 ms.
| `with the number. Any GitHub closing keyword works: Closes / Fixes / Resolves, optional colon, #N or a full issue URL${tail}`, | ||
| ); | ||
| } | ||
| if (linked.length > 1) { |
There was a problem hiding this comment.
When a body contains Closes #12 and Closes other/repo#4, this condition sees only the local linked array and passes; if the author has permission in the other repository, merging can therefore close two issues despite the documented one-issue-per-PR contract. The prior fix still excludes the separately collected cross references, so reject additional cross-repository closing references or count all parsed references here.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 31c13d1: any cross-repository closing reference is now a failure pointing at rule 5 with the suggestion to use Refs; the cross-only and cross-plus-local cases are both tested.
…a second issue Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 31c13d1ec5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| async function verifyIssue(github, owner, repo, number) { | ||
| let issue; | ||
| try { | ||
| ({ data: issue } = await github.rest.issues.get({ owner, repo, issue_number: number })); |
There was a problem hiding this comment.
In a private repository created from this template, .github/workflows/ci.yml grants the GITHUB_TOKEN only contents: read, which sets unspecified scopes to none. The new caller reaches this issues.get request, but reading issues in a private repository requires issues: read; consequently every otherwise-valid, nonexempt PR throws could not verify issue and cannot satisfy the sole required ci check. Grant issues: read to this job or workflow.
AGENTS.md reference: AGENTS.md:L19-L19
Useful? React with 👍 / 👎.
| .replace(/(^|\n)(`{3,}|~{3,})[^\n]*\n[\s\S]*?\n\2[ \t]*(?=\n|$)/g, '$1') | ||
| .replace(/`[^`\n]*`/g, ''); |
There was a problem hiding this comment.
When the only closing keyword is in a four-space-indented code block, a fence indented by up to three spaces, or an unterminated fence, GitHub renders it as code and does not create a closing reference, but this replacement handles only closed fences beginning in column zero, so linkedIssues() accepts the keyword and the required check goes green. Fresh evidence in the current implementation is that bodies such as Closes #42 and ```\nCloses #42 are still accepted after the earlier fenced-code fix; parse the Markdown or cover these valid block forms as well.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
| if (issue && issue.pull_request) return `#${number} is a pull request, not an issue — link the issue the work is for`; | ||
| return null; |
There was a problem hiding this comment.
When the linked number belongs to an already-closed issue, issues.get succeeds and this function returns no problem, so the PR passes even though merging it cannot close that issue. This defeats the newly documented skills/pr-authoring/SKILL.md rule 5 contract that every PR closes one issue; inspect the returned issue state and require an open issue, with guidance to reopen it or create the appropriate follow-up issue.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in c6c5895: state === 'closed' fails with 'reopen it, or open a follow-up issue and link that'.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c6c5895e2e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const result = lint({ | ||
| headRef: pr.head && pr.head.ref, |
There was a problem hiding this comment.
When a PR targets a feature or release branch rather than the repository's default branch, GitHub does not apply closing keywords, but the linter only passes the head ref and body and therefore reports success after verifying the issue. Such a PR can merge without closing or linking its claimed issue; inspect pr.base.ref against pr.base.repo.default_branch before accepting the reference.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
| return String(text || '') | ||
| .replace(/<!--[\s\S]*?-->/g, '') | ||
| .replace(/(^|\n)(`{3,}|~{3,})[^\n]*\n[\s\S]*?\n\2[ \t]*(?=\n|$)/g, '$1') | ||
| .replace(/`[^`\n]*`/g, ''); |
There was a problem hiding this comment.
Fresh evidence after the earlier inline-code fix is that a body containing only Closes #42 inside a valid two-backtick code span is reduced to plain Closes #42 by this single-backtick regex and passes the lint, although GitHub treats it as code and will not close the issue. Strip code spans using their complete matching delimiter length, or parse the Markdown, before scanning for closing keywords.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 7df8355: code spans are now stripped by a linear scanner that closes a run of N backticks only with a run of exactly N (CommonMark), and fences by a line scanner (unclosed fence runs to EOF; a backtick fence whose info string contains a backtick is not a fence). Tests: Closes #15, `Closes #16` + prose, triple-backtick span + prose. Adversarial 65k bodies (all-backtick, alternating, 16k fence openers) each under 2 ms.
…ck runs by length Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7df8355021
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
|
|
||
| function stripNonProse(text) { | ||
| return stripCodeSpans(stripFencedBlocks(String(text || '').replace(/<!--[\s\S]*?-->/g, ''))); |
There was a problem hiding this comment.
When a body contains an unterminated comment such as <!--\nCloses #42, CommonMark treats the comment as continuing to EOF, so the closing keyword is not prose and GitHub will not close the issue. This replacement only removes comments with a terminating -->, causing linkedIssues() to accept the hidden reference and let the required check pass; strip from an unmatched opener through EOF as well.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in b872a9c: an unterminated <!-- that starts a line (≤3 spaces) is an HTML block to EOF and is dropped; unterminated mid-line it is literal text, which GitHub renders and links, so it is kept — both tested. The comment strip is now a linear scanner; every adversarial 65k body (16k openers, 13k mid-line openers, 8k closed pairs, backtick runs, 16k fences, padded headings) finishes in under 25 ms.
…inear Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b872a9c709
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
|
|
||
| function stripNonProse(text) { | ||
| return stripCodeSpans(stripFencedBlocks(stripHtmlCommentsLinear(String(text || '')))); |
There was a problem hiding this comment.
When a fenced code block contains a line-start <!-- and the body has its real Closes #42 after that fence, stripHtmlCommentsLinear() runs first and treats the opener as an unterminated HTML block through EOF. GitHub/CommonMark instead treats it as literal fenced-code content, so the later closing reference is valid, but this linter removes it and blocks the PR as linking no issue. Account for Markdown construct nesting rather than stripping comments before recognizing their enclosing code fences.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 3f4f902: one line-by-line block pass first (fence or HTML block, whichever opens first, each with its own close rule), then inline code spans, then inline comments — so <!-- inside a fence is code, ``` inside a comment block is comment, and a backtick left of <!-- makes it code. Six nesting cases added to the tests; worst adversarial 65k body 22 ms.
…ark precedence) Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3f4f902ea0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const BRANCH_RE = new RegExp(`^(${TYPES.join('|')})/(\\d+)-[a-z0-9][a-z0-9._-]*$`); | ||
| // GitHub's closing keywords: optional colon, then `#N`, `owner/repo#N`, or a | ||
| // full issue URL — the three forms GitHub itself links and auto-closes. | ||
| const CLOSES_RE = /\b(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?):?\s+(?:https?:\/\/github\.com\/([\w.-]+\/[\w.-]+)\/issues\/(\d+)|([\w.-]+\/[\w.-]+)?#(\d+))\b/gi; |
There was a problem hiding this comment.
When a body contains Closes, then a fenced code block, then #42, stripBlocks() removes the entire fence and this \s+ matches the remaining text across the block boundary. GitHub renders the keyword and reference in separate prose/code blocks, so they do not form a closing phrase; if issue #42 is open, API verification still succeeds and CI goes green even though merging will not close it. Preserve a non-whitespace boundary for removed blocks and avoid matching across paragraph boundaries.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
| # is here — an adopter who takes a thin-caller workflow without its script | ||
| # gets a job that fails on every event while `make check` stays green. | ||
| missing=0 | ||
| for wf in .github/workflows/*.yml; do |
There was a problem hiding this comment.
When an adopter adds a standard .github/workflows/*.yaml workflow that requires a scripts/*.js handler, this loop never examines it because it only expands *.yml. A missing handler can therefore leave make check green while the workflow fails on every event, defeating the new completeness check; scan both .yml and .yaml files.
AGENTS.md reference: AGENTS.md:L19-L19
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 18a4509: both suffixes scanned (the [ -f ] guard already skips an unexpanded glob).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 18a4509e25
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if [ ! -f "$js" ]; then | ||
| echo "FAIL: $wf requires $js, which is missing — take the workflow and the script together (docs/template/upgrading.md)" | ||
| missing=$((missing + 1)) | ||
| else | ||
| echo "OK: $wf requires $js (present)" |
There was a problem hiding this comment.
When an adopter adds a workflow that requires an existing scripts/foo.js but omits scripts/foo.test.js, this branch only confirms that the handler exists, and the later node --test scripts/*.test.js command passes by running the repository's other tests. make check therefore stays green even though the new event handler is completely untested; verify a sibling test exists for each workflow-required script before running the suites.
AGENTS.md reference: AGENTS.md:L17-L19
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 26354b3: a required scripts/X.js without scripts/X.test.js fails naming ADR-0008; fixture verified.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 26354b3554
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (m - k === n) { closed = true; break; } | ||
| k = m; | ||
| } | ||
| if (closed) { i = k + n; continue; } |
There was a problem hiding this comment.
When a body contains Closes followed by an inline code span and then #42, this branch deletes the span entirely, so Closes ignored #42 becomes Closes #42 and passes lint after the API lookup. GitHub renders the code node between the keyword and reference, so they do not form a closing phrase and merging will not close the issue; replace removed spans with a non-whitespace sentinel or otherwise prevent matching across them.
AGENTS.md reference: AGENTS.md:L50-L50
Useful? React with 👍 / 👎.
| echo "OK: $wf requires $js (present, tested by ${js%.js}.test.js)" | ||
| fi | ||
| done <<EOF_JS | ||
| $(grep -oE 'scripts/[A-Za-z0-9._-]+\.js' "$wf" | sort -u) |
There was a problem hiding this comment.
When a workflow merely mentions scripts/foo.js in a comment or invokes a non-handler script, this unrestricted grep still classifies it as a required event handler and fails make check unless scripts/foo.test.js exists. That makes harmless workflow documentation—or scripts outside ADR-0008's thin-caller pattern—block validation; extract paths only from the workflow's actual require(...) calls.
AGENTS.md reference: AGENTS.md:L19-L19
Useful? React with 👍 / 👎.
## Summary `skills/branch-and-commit`'s type table said `refactor` produces a patch bump — release-please hides it, no release. It also said `ci` produces no bump, which #69 ignored: the PR lint shipped as `ci:` and release-please skipped it (`No user facing commits found`). Both skills now say what to do when a release-worthy change only fits a hidden type. This PR is the recipe in use. ## Related issue Closes #70 ## Validation - [x] L0 static — `make verify` green - [x] L3 — this PR's merge must produce `chore: release 0.5.2` (the footer below) ## Risk / rollback ```text RISK: none — documentation. Rollback: revert. ``` ## Checklist - [x] Conventional Commit PR title - [x] Linked issue using "Closes #N" - [x] No secrets, no `*.local.md` files committed - [x] Documentation updated where affected Release-As: 0.5.2 Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
1 participant
Summary
Six downstream PRs were opened with a bare
Closes #and achore/-slugbranch and nothing rejected them.scripts/pr-lint.js(ADR-0008 shape, with tests) now checks: the head branch is<type>/<issue#>-<slug>(types = the list inAGENTS.md's Branch step, which is now that list's one home — the test asserts equality); the body links an issue in this repository with a GitHub closing keyword (colon form,owner/repo#Nand full-URL forms recognised; HTML comments stripped first, so the untouched template line fails); the branch's number is among the linked ones. release-please and Dependabot branches are exempt.ci.ymlruns it as a step-level-ifgithub-scriptstep onpull_request, whosetypes:now also includeeditedandready_for_review(a superset of the defaults — header amended to forbid anything less) so a fixed body turns the check green without a push.scripts/check-node-tests.shreplacescheck-issue-labeler.sh, runs everyscripts/*.test.js, and fails when any workflowrequires ascripts/*.jsthat is missing. The pr-authoring recipe no longer submits the template verbatim.Related issue
Closes #56
Validation
make verifygreen (actionlint on the widenedon.pull_request, checks incl. 26 node tests)scripts/pr-lint.test.js: 12 cases (the six-PR failure shape, untouched template line, comment-hidden keyword, all keyword forms incl.Closes: #Nand URLs, cross-repo kept apart, Refs-only guidance, branch grammar for every type and eight rejects, bot exemption, run() fail/pass/skip)chore/-initial-version-docsPRs fail on both counts, their redonechore/61-…/chore/12-…pass. Torn-upgrade fixture (ci.ymlthin caller,pr-lint.jsabsent) →make checkFAIL naming the file.edited-event gap, the colon form, the recipe that would have failed its own lint, and the type list living in three disagreeing places.cirun executes the new step on its own branch/body (ci/56-pr-lint,Closes #56).Risk / rollback
Checklist
*.local.mdfiles committed