メニュー
Full Phoenix React performance workflow with git branch, commit, and optional PR. Wraps /phoenix-perf-hunt with git automation.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Implement, debug, test, and review changes in Phoenix IDE. Use for Phoenix coding tasks, repository workflow, validation, codegen, commits, or choosing the right specialized Phoenix skill. Start here for Work-mode tasks; use phoenix-explore first when the problem is still ambiguous.
Breadth-first, read-only investigation of ambiguous Phoenix IDE feedback and requests. Use in Explore mode to trace a raw symptom across easy-to-miss UI, runtime, wire, persistence, tool, worktree, and production boundaries before proposing an implementation task. Produces a bounded evidence-backed task, not code changes.
Production deployment for Phoenix IDE. Use when running ./dev.py prod deploy, checking production status, diagnosing a failed deploy, or stopping the production service.
Cut a new phoenix-ide version by merging a version-bump PR, which triggers GitHub Actions to tag, build, and publish the binary, then replace the auto-generated release notes with a sub-agent-drafted, human-reviewable changelog. Use when the user says "cut a release", "publish a version", "ship vX.Y.Z", "tag a new release", or asks how to publish.
Finds one valid React performance target in the Phoenix UI. Profiles scenarios via the run-scenario harness (default transport: browser_profile), scans for known React anti-patterns, learns from db.yaml history, ranks, and returns a filled target.yaml. Use before /phoenix-perf-hunt.
Coordinates a Phoenix React performance attempt. Captures the scenario baseline (raw samples) BEFORE any change via the run-scenario harness (default transport: browser_profile), implements one focused fix, hands off to review, and records every outcome in db.yaml. Does not judge.
| name | phoenix-perf-submit |
| description | Full Phoenix React performance workflow with git branch, commit, and optional PR. Wraps /phoenix-perf-hunt with git automation. |
| allowed-tools | Bash Read Skill |
Wraps /phoenix-perf-hunt with git automation. Push/PR require explicit user
authorization (per AGENTS.md): commits are local and fine; pushes affect others.
Run /phoenix-perf-preflight.
git status
STOP if the working tree is dirty. Baseline must be captured against a known clean commit, else the post-change diff is not exactly the one change under test. Commit/stash unrelated work first.
Run /phoenix-perf-hunt. After it completes, return here. Hunt records the
verdict in db.yaml; it does not branch, commit, push, or PR. Those are this
skill's job.
If hunt recorded blocked (scenario unrunnable): STOP. Nothing to submit —
there is no measured improvement. Report the blocker to the user.
If hunt recorded rejected: STOP. Do not commit a rejected change; revert it
(git checkout -- <file>), leave the db record (it teaches future hunts).
git checkout -b perf/ui-<technique>-<short>
git add -A
Commit using resources/commit-template.txt. First line ≤ 50 chars. Put real
measured numbers (from review's report) in the body — no estimates.
git push -u origin perf/ui-<technique>-<short>
gh pr create --title "perf(ui): <short>" --body "$(cat <<'EOF'
## Summary
<what re-render / scripting cost was removed and why>
## Measured (run-scenario / browser_profile, scenario <name>, N=<n>)
- script_ms: <base> -> <opt> (<-%>, p=<p>)
- react_commits: <base> -> <opt> (<-%>, p=<p>)
- react_actual_ms: <base> -> <opt> (<-%>, p=<p>)
- js_heap_used_mib: <base> -> <opt> (<-%>, p=<p>)
## Validation
- [x] ./dev.py check passes
- [x] 5-persona review unanimous (db: skills/phoenix-perf-hunt/resources/db/<id>.yaml)
- [x] No behavior change (rendered output / SSE / draft isolation unchanged)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
EOF
)"