Post pr review

Skill danielcg-net/claude_status_dashboard/.claude/skills/post-pr-review

Post-PR review loop — monitor DeepSeek Code Review comments, classify issues vs false positives, ensure CI passes, iterate fixes. Invoke after a PR is pushed or when the user says "review the PR," "check CI," "handle DeepSeek comments," "PR review loop."From its SKILL.md

Install

npx -y skills add danielcg-net/claude_status_dashboard --skill post-pr-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.4 KB, as published. Nobody here has run it

Post-PR Review Loop

After a PR is pushed, two things must happen before merge: (1) the DeepSeek Code Review GitHub Action posts review comments, and (2) all required CI checks must pass.

The Loop

┌─────────────────────────────────────────────────┐
│  1. Wait for DeepSeek CR check to complete       │
│  2. Fetch all PR review comments                  │
│  3. Classify each: relevant / false-positive      │
│  4. Fix real issues, reply to false positives     │
│  5. Check CI — all required checks green?         │
│  6. If issues remain or CI red → loop to step 1   │
│  7. PR is ready to merge                          │
└─────────────────────────────────────────────────┘

Step 1 — Wait for DeepSeek CR

Use mcp__plugin_github_github__pull_request_read with method=get_check_runs. Wait until "deepseek-code-review" is completed.

Step 2 — Fetch Review Comments

Fetch both review comments and issue comments from the PR, filtering to the github-actions[bot] author.

Step 3 — Classify Each Finding

For each DeepSeek comment:

Read the actual code at the flagged line — never trust the comment alone
Check against known false-positive patterns from AGENTS.md and the deepseek-cr.yml system prompt
Classify: Real issue / False positive / Style nit

Known false positives for this project:

|| in asNumber() patterns in src/usage.ts — intentional ccusage fallback
createElement string children — not innerHTML, these are text nodes
Duplicate types in client.ts vs usage.ts — separate bundles, intentional
No Zod validation for usage data — usage.ts parses external ccusage data

Step 4 — Address Findings

Real issues: Fix, commit, push. False positives: Reply explaining why, resolve thread. Style nits: Skip unless contradicting AGENTS.md.

Step 5 — Check CI

Required checks:

Check	Source
Type check and unit tests	`test.yml`
DeepSeek CR	`deepseek-cr.yml`

All must be green. If Type check and unit tests fails, investigate the build logs. Assume it's your fault until proven otherwise.

Step 6 — Loop or Finish

Pushed fixes → go back to Step 1
CI red → investigate, fix, push
All green, all comments resolved → merge

Analysis Heuristics

Does the code actually do what the comment claims? Read the file.
Would fixing this break an established project convention? Check AGENTS.md.
Is the flagged line reachable? DeepSeek sometimes flags test fixtures.
Does the fix make the code objectively better? Skip neutral changes.

Gotchas

DeepSeek comments arrive after the check completes — wait 30s before fetching
Force-pushing after review creates stale line references — add new commits instead
continue-on-error: true on DeepSeek job means green check ≠ no errors