메뉴
Use when the user wants to replay a Claude Code session from a GitHub Actions artifact link. Downloads the artifact, finds JSONL files, generates an interactive HTML replay with claude-replay, and serves it locally.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
PR review orchestrator. Triages the change, dispatches specialized sub-agents in parallel across review dimensions, synthesizes their findings, runs PR-specific checks, and produces a structured review result. Sub-agent definitions live in sub-agents/ relative to this file.
Check whether ADR numbers in the current branch conflict with ADRs already on the PR's target branch, and renumber if needed. Use before merging a PR that adds new ADRs, to avoid number collisions with concurrently merged ADRs.
Use when the user wants to tag a release, cut a release candidate, or ship a new version. Also use when asking about release process, versioning, or how GoReleaser is configured.
Standalone procedure for reviewing any code change. Identifies the change, reads surrounding source, evaluates across six review dimensions, and compiles structured findings. Can be invoked directly for local review or delegated to by the pr-review skill.
Use when checking e2e test health or reviewing recent e2e failures on main.
Use when you need to analyze the delta between agent review and human review on a PR to identify structural repo improvements that would close review gaps or justify increased agent autonomy.
| name | replay-session |
| description | Use when the user wants to replay a Claude Code session from a GitHub Actions artifact link. Downloads the artifact, finds JSONL files, generates an interactive HTML replay with claude-replay, and serves it locally. |
| compatibility | Requires claude-replay (npm install -g claude-replay) and gh CLI authenticated with access to the target repository |
Generate an interactive HTML replay from a JSONL transcript stored in a GitHub Actions artifact.
claude-replay installed globally (npm install -g claude-replay)gh CLI authenticated with access to the target repositoryThe user provides a GitHub Actions run URL. Extract the owner, repo, and run ID from it.
URL formats to expect:
https://github.com/OWNER/REPO/actions/runs/RUN_IDhttps://github.com/OWNER/REPO/actions/runs/RUN_ID/attempts/Nhttps://github.com/OWNER/REPO/actions/runs/RUN_ID/artifacts/ARTIFACT_IDBefore downloading, interactively ask the user:
gh run view <run-id> -R OWNER/REPO to list the artifacts available in that run. If there are multiple artifacts, ask which one to use.TMPDIR=$(mktemp -d /tmp/replay-XXXXXX)
gh run download <run-id> -R OWNER/REPO -n "<artifact-name>" -D "$TMPDIR/artifact"
The downloaded artifact represents a single agent run. Its top-level directory (e.g., agent-code-2446-1776439672/) is the agent identifier. All JSONL files inside — across all iteration-N/ subdirectories — belong to this agent. There is no need to ask the user which agent to replay.
find "$TMPDIR/artifact" -name "*.jsonl" -type f
List the files to the user grouped by iteration, with sizes, so they can choose "all" (default) or a specific subset.
claude-replay <path-to-jsonl> --no-compress -o "$TMPDIR/replay.html" [extra options]
claude-replay <path-to-jsonl> --no-compress --serve --port 7331
Always use --no-compress — compressed replays fail to render in some browsers.
If port 7331 is busy, try 7332, 7333, etc.
Present the URL to the user: http://127.0.0.1:<port>
If the user wants to replay multiple files (e.g., main + subagents), chain them:
claude-replay file1.jsonl file2.jsonl file3.jsonl -o "$TMPDIR/replay.html"
/tmp for downloaded artifacts and generated HTML — never pollute the working directory.--serve flag works even though the process appears to exit immediately in the background. The server stays up — don't retry or switch to a different HTTP server.kill $(lsof -ti:<port>) 2>/dev/null