pr-cycle

name	pr-cycle
description	Full PR review cycle — fetch open review comments, spawn an independent code-review sub-agent and a library-consumer sub-agent, evaluate all findings, apply valid fixes, run CI, amend and push, resolve all threads, and re-request review from Copilot. Use when asked to "run the pr cycle", "address pr comments", "resolve comments and re-request review", or after pushing a new round of changes.

PR Cycle

Run one full iteration of the review → fix → push → re-request loop.

Input

Optional PR number. If omitted, infer it from the current branch using gh pr view --json number.

Steps

1. Fetch open Copilot comments

Run outside the sandbox (GitHub API requires network):

gh api repos/OWNER/REPO/pulls/PR/comments --paginate

Filter to comments created after the last push (compare created_at to the last force-push timestamp). Print each with: ID, file:line, and body.

Also fetch unresolved review threads via GraphQL to get their node IDs for later resolution:

gh api graphql -f query='{ repository(owner:"OWNER", name:"REPO") { pullRequest(number:PR) { reviewThreads(first:50) { nodes { id isResolved comments(first:1) { nodes { databaseId } } } } } } }'

2. Spawn two independent sub-agents in parallel

Agent A — code reviewer: Spawn with the code-reviewer subagent type (or general-purpose if unavailable). Prompt must include:

• The PR number and branch name
• The full diff (git diff origin/master)
• Instructions to review for correctness, API design, test coverage, documentation accuracy, and Rust idiom adherence
• Instructions to flag any issues with severity (high / medium / low)
• Instructions NOT to fix anything — report only

Agent B — library consumer: Spawn a general-purpose agent in a read-only research role. Prompt must include:

• The PR number and branch name
• The full diff (git diff origin/master)
• The current src/lib.rs doc comments and README.md (or the relevant excerpts covering the changed APIs)
• Instructions to evaluate the changes solely from the perspective of a downstream crate author who is adding or upgrading cached as a dependency — not as a reviewer of the implementation
• Specifically ask it to assess:
  • Is the public API surface intuitive? Are names, method signatures, and trait bounds what a user would expect?
  • Are the docs sufficient to use the feature without reading the source? Are there gaps, ambiguities, or missing examples?
  • Are there footguns — easy mistakes a user could make that aren't warned about in the docs?
  • Does the feature compose naturally with existing cached features (e.g., result, option, result_fallback, sync_writes)?
  • Are error messages from compile-fail cases clear enough for a user to self-diagnose?
• Instructions to flag any issues with severity (high / medium / low)
• Instructions NOT to fix anything — report only

Launch both agents in parallel. Wait for both to complete before proceeding.

3. Evaluate all findings

Present all three sets of findings (Copilot comments + code-reviewer + consumer reviewer) together. For each finding:

• Valid: the concern is real and the code should change
• Invalid: the finding is incorrect, already addressed, or environment-specific (e.g. rustc version mismatch on trybuild golden files)

Explain your reasoning for each verdict. Do not apply any fix silently — call out what you are doing and why.

4. Apply fixes for valid findings

For each valid finding, make the minimal correct fix. Common fix types for this repo:

• Documentation/comment updates in src/stores/, src/lib.rs, or cached_proc_macro/src/lib.rs
• Test additions or corrections in tests/cached.rs
• Trybuild golden file regeneration: TRYBUILD=overwrite cargo test --no-default-features --features "proc_macro,time_stores" compile_fail_macro_arg_validation
• Macro code changes in cached_proc_macro/src/

5. Run CI

make ci

Fix any errors. Redis and Docker failures are expected in local environments and can be ignored. All other failures must be resolved.

If trybuild golden files drift, regenerate them:

TRYBUILD=overwrite cargo test --no-default-features --features "proc_macro,time_stores" compile_fail_macro_arg_validation

6. Regenerate README if src/lib.rs changed

cargo readme --no-indent-headings > README.md

7. Amend and push

git add -p   # stage only changed files explicitly
git commit --amend --no-edit
git push --force-with-lease origin BRANCH

Do not add a Co-Authored-By line.

8. Resolve all open threads

For each unresolved thread node ID collected in step 1:

gh api graphql -f query='mutation { resolveReviewThread(input: {threadId: "THREAD_ID"}) { thread { id isResolved } } }'

Run outside the sandbox. Do not leave replies — resolve silently.

9. Re-request Copilot review

gh api repos/OWNER/REPO/pulls/PR/requested_reviewers \
  -X POST -f 'reviewers[]=copilot-pull-request-reviewer[bot]'

10. Report

• How many Copilot comments were found, how many were valid, how many were fixed
• How many code-reviewer findings were found, how many were valid, how many were fixed
• How many consumer-reviewer findings were found, how many were valid, how many were fixed
• Which findings were ruled invalid and why
• Confirm threads resolved and Copilot re-requested
• The resulting commit SHA and push status