PR Review

Reviews open PRs on sunitghub/canon-skills. Human decides Fix Y/N — skill handles mechanics.

Steps

1. Fetch open PRs

gh pr list --repo sunitghub/canon-skills --state open \
  --json number,title,body,url,createdAt

If no open PRs: report "No open PRs on sunitghub/canon-skills." and stop.

2. Read each diff

For each open PR:

gh pr diff <number> --repo sunitghub/canon-skills

3. Read current review file

Read PRs/canon-skills-pr-review.md. Extract the PR numbers already in the table (first column) — these are existing entries to update in place, not duplicate.

4. Analyse and present

For each finding you'd otherwise raise, score it 0-100 before including it:

• 0 — false positive: doesn't hold up, or a pre-existing issue not introduced by this PR.
• 25 — unverified or a stylistic nit not called out by any convention in this repo.
• 50 — verified real, but a minor nitpick, low-impact in practice.
• 75 — verified real and important: will affect functionality, correctness, or maintainability in practice.
• 100 — certain and will be hit frequently in practice.

Drop anything scoring below 80 — do not include it in the findings block below. Check each candidate finding against this false-positive list before scoring:

• Pre-existing issue, not introduced by this PR
• Looks like a bug but isn't
• Pedantic nitpick a senior engineer wouldn't raise
• Something a linter, typechecker, or compiler would already catch
• Silenced via an explicit lint-ignore comment
• On a line the PR didn't modify

For each PR, produce a one-block summary from the findings that survive the filter:

PR #<n> — <title>
Area:     <one of: Tooling / DX / Sprint / MCP / Docs / Meta>
Findings: <2–4 sentences: what it does, what's good, what's risky or wrong>
Recommend: Fix Y / Fix N — <one-line reason>

Present all PRs, then ask:

"Fix Y/N for each? I'll wait for your call before updating the review file or creating tickets."

5. Receive decisions

Wait for the user to confirm Y/N per PR. Accept any clear format ("Y on #2, N on #3", "#2 yes", etc.). Do not proceed until all open PRs have a decision.

6. Update PRs/canon-skills-pr-review.md

Table columns (preserve existing order): | PR | Title | Area | Findings | Verdict | Opened | Reviewed |

Date format:

• Opened: PR createdAt converted to Central time and formatted as MM/DD/YYYY.
• Reviewed: date this review pass updated the row, formatted as MM/DD/YYYY.
• Existing entry (PR number already in table): update the row in place — replace Findings, Verdict, and Reviewed with the current analysis/date. Preserve Opened unless it is missing or in an older format; normalize old Opened values to MM/DD/YYYY. Do not add a new row.
• New entry: insert a new row at the top of the table (below the header), above existing rows. Set Verdict to Merge (Fix=Y) or No (Fix=N), Opened from createdAt, and Reviewed to today's review date.
• Escape any | inside cell text as \|.

7. Create tickets for Fix=Y PRs

For each PR where the user confirmed Fix=Y, run:

tkt create "implement: <short title referencing PR #n>"

Then open the created ticket's ticket.md and append to the description body:

Source: <PR URL>
Area: <area>
Findings: <findings summary>

Report the ticket ID created for each PR.

8. Draft and post comments

For each PR (Y or N), draft a comment summarising the outcome. Cite each specific finding with a full-SHA blob link, not just prose — this lets the PR author jump straight to the flagged code:

https://github.com/sunitghub/canon-skills/blob/<full-sha>/<path>#L<start>-L<end>

Fetch the full head SHA for the link (do not use git rev-parse or any local shell substitution — the link must render as literal markdown, and this flow never checks out the PR branch):

gh pr view <number> --repo sunitghub/canon-skills --json headRefOid -q .headRefOid

Include at least 1 line of context before and after the flagged line in the range (e.g. a finding on line 12 links #L11-L13).

Fix=Y template:

Thanks for the PR. After review:

**Area:** <area>
**Findings:** <findings, with a blob link per cited issue>
**Verdict:** Merging — filed as <ticket-id> to track implementation.

Fix=N template:

Thanks for the PR. After review:

**Area:** <area>
**Findings:** <findings, with a blob link per cited issue>
**Verdict:** Not merging as-is — <one sentence on why / what would change the verdict>.

Show all drafts and ask:

"Post these comments? (yes / edit first / skip)"

On yes: immediately before posting each comment, re-check the PR is still eligible — it may have been closed, merged, or converted to draft since step 1:

gh pr view <number> --repo sunitghub/canon-skills --json state,isDraft

If state is no longer OPEN or isDraft is now true, skip posting that comment and note it in chat instead. Otherwise post via:

gh pr comment <number> --repo sunitghub/canon-skills --body "<draft>"

On edit first: show the draft for the specific PR, accept the edit, reconfirm, then post (with the same eligibility re-check).

On skip: do not post. Note it in chat.

Constraints

• Never post a comment without explicit user approval.
• Never create a ticket without a Fix=Y decision from the user.
• Do not duplicate rows in the review file — match by PR number.
• Scope analysis to the PR diff only. Do not re-read the full canon codebase unless a specific file is directly relevant to a finding.