You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
path_instructions cover scripts/** and tests/** only. Nothing covers docs/** (seven files after the README split), .github/** (where the
release automation and a write-capable token live), statusline/**, or hooks/**.
Proposed behavior
Raise reviews.profile to assertive.
Add path instructions for the uncovered paths, each carrying the conventions
that actually apply there:
docs/**: every documented command must exist, anchors must resolve, and no
other statusline tool is named.
.github/**: actions pinned to a full commit SHA, and a workflow holding a
write-capable token must not execute a script from the pull request's own
revision.
statusline/** and hooks/**: never fail the session — a missing
dependency or unreadable config prints nothing and exits 0.
Consider auto_assign_reviewers: true so its review is visible as a review.
Alternatives considered
Leaving chill and requesting a full review by hand on any branch that
touches something delicate. Works — it is what caught the statusLine.type
defect — but it depends on remembering.
Raising the profile only for some paths. Not supported; profile is global.
Additional context
Surface: .coderabbit.yaml. No code change, no release.
Contrary to a reasonable first assumption, "turn on more linters" is already
done: actionlint, yamllint, markdownlint and gitleaks all default to
enabled, and listing only shellcheck under tools: has not disabled them.
Worth confirming they are actually firing, since the docs-only #32 drew no
markdown comments at all.
Problem
.coderabbit.yamlhas served four feature branches now, and its gaps arevisible in what got reviewed and what did not.
profile: chill. At that level, feat(doctor): report a CLI older than the plugin, and a statusLine that is not ours #33 needed a second, explicitly requestedfull review to surface two defects the incremental pass had missed — one of
them a classifier that ignored
statusLine.typeand so reported a brokenconfiguration as correct.
path_instructionscoverscripts/**andtests/**only. Nothing coversdocs/**(seven files after the README split),.github/**(where therelease automation and a write-capable token live),
statusline/**, orhooks/**.Proposed behavior
reviews.profiletoassertive.that actually apply there:
docs/**: every documented command must exist, anchors must resolve, and noother statusline tool is named.
.github/**: actions pinned to a full commit SHA, and a workflow holding awrite-capable token must not execute a script from the pull request's own
revision.
statusline/**andhooks/**: never fail the session — a missingdependency or unreadable config prints nothing and exits 0.
auto_assign_reviewers: trueso its review is visible as a review.Alternatives considered
chilland requesting a full review by hand on any branch thattouches something delicate. Works — it is what caught the
statusLine.typedefect — but it depends on remembering.
profileis global.Additional context
Surface:
.coderabbit.yaml. No code change, no release.Contrary to a reasonable first assumption, "turn on more linters" is already
done:
actionlint,yamllint,markdownlintandgitleaksall default toenabled, and listing only
shellcheckundertools:has not disabled them.Worth confirming they are actually firing, since the docs-only #32 drew no
markdown comments at all.