Repository navigation
Conversation
Every later change to the review workflow changes who can cause a review, so each rule needs a test that runs before the change merges. The request decision reads a GitHub event and decides whether to send it on as a review request. Its tests are saved event payloads, each with the decision expected for it, run by a bash and jq script that needs no network or GitHub runner, so the same tests run on a laptop and in CI. For now the decision only skips events that are not a kind of review request; the kinds themselves come in later tickets. actionlint runs through `go run` at a pinned version, so Go's module checksum database verifies what runs. Closes #2 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Jmn7ot2xooXpcuW44QqtU
8 tasks
cipherstash-agent
marked this pull request as ready for review
November 17, 2026 23:54
This was referenced Nov 18, 2026
Open
…o cases The test script stopped at the first case whose expected.json was missing or not valid JSON, so the cases after it never ran and the summary never printed. A run that found no cases at all passed, so a moved or emptied cases directory would turn CI green without testing anything. A missing or invalid expected.json now counts as a failed case and the run carries on. A run with no cases fails. request-decision/test-runner checks these rules against copies of the scripts, and CI runs it. The README and the workflow header now say the checks also run on every push to main, and one README line holding two sentences is split. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Jmn7oR2GLajZ1RUvvkpdA
…sion The checks of request-decision/test copied the real decision, which reads decide.jq here but reads its program out of the reusable workflow further up the stack. A stand-in that echoes its input keeps the checks about the test script alone, so they run unchanged on every branch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Jmn7oR2GLajZ1RUvvkpdA
tobyhede
approved these changes
Nov 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This adds the checks every later change to the shared review workflow depends on.
Every pull request to this repository, and every push to
main, now lints the workflows and runs the request decision against saved GitHub event payloads.The request decision is the part that reads a GitHub event and decides whether it is a review request to send on to Claude.
For now it only skips events that are not a kind of review request.
Changes
main. It has two jobs:actionlint, and the request decision's job, which checks the test script and then runs the saved cases. It has read-only permissions, the checkout is pinned by full SHA, and actionlint is pinned to v1.7.12 and verified by Go's module checksum database.jqprogram and a shell wrapper. It returns{"decision": "skip", "reason": "not-a-review-request"}for every input.request-decision/testpipes each case'sinput.jsoninto the decision and compares the result with the case'sexpected.json. It needs onlybashandjq. A case with a missing or invalid file counts as a failed case, and the run goes on to the next case. A run that finds no cases fails. There is one case: anissuesevent is skipped.request-decision/test-runnerchecks these rules forrequest-decision/testagainst a stand-in decision, so the check does not depend on what the real decision does.The first commit on
main(agent docs, glossary, licence and README) was pushed directly. The repository was empty, and a pull request needs a base branch.Verification
request-decision/test: 1 case, 0 failed.request-decision/test-runner: 5 checks pass. A passing case passes. No cases fails. A missingexpected.json, an invalidexpected.json, or a missinginput.jsoneach count as one failed case. On the test script from before the fix, 3 of these checks failed.bash -nfinds no syntax errors in the scripts.actionlintandshellcheckwere not run locally, because they are not available here. The Checks run on this pull request runs them.Related
Review notes
One thing for #5 to settle, not this pull request: a reusable workflow runs in the calling repository, so the request decision's files are not on disk there.
#5 needs the reusable workflow to fetch them from this repository at the same commit the caller pinned, or to carry the
jqprogram inline.The fixes for the earlier review's suggestions (missing or invalid
expected.json, no cases, thepushtrigger) were merged up into #11, #12 and #13 with merge commits.Generated by Claude Code