Menu
Run long local commands with live combined output, persistent logs, idle reporting, exit-code propagation, and distilled JSON evidence. Use for builds, tests, or other verbose tasks that need monitoring.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | task-watch |
| description | Run long local commands with live combined output, persistent logs, idle reporting, exit-code propagation, and distilled JSON evidence. Use for builds, tests, or other verbose tasks that need monitoring. |
Run the command after a required literal -- separator without a shell. Never omit the separator; it prevents watcher options from being confused with child arguments.
python3 scripts/task_watch.py --log build.log --json-summary build.json -- make test
Tune --idle-seconds, --heartbeat-seconds, and --summary-lines as needed. Idle detection reports state and does not kill the child. The JSON summary preserves the command, outcome, timing, log location, and numbered error/fail/warn-plus-tail evidence.
Forward output in chunks so flushed text updates activity even before a newline. Keep the queue and evidence collector bounded by fixed chunk and summary limits; only the disk log retains complete output.
Use this skill to absorb large local output and hand back concise evidence. It complements ci-monitoring, which follows remote CI rather than local long-running tasks.
Generates a Context Snapshot (Why/rationale + file changes) before /compact to preserve important decision history across Claude Code sessions. Use when running /compact or when you need to save session context. Hooks auto-intercept both manual and auto-compact.
Extract keyframes from MP4/GIF/MOV video files into a single strip image for Claude to analyze. Use when the user shares a video file and wants to understand animation behavior, UI transitions, or visual bugs that are hard to describe with static screenshots.
Validate commit message structure, required Codex co-author attribution, and staged-path breadth. Use before creating a Git commit or reviewing commit readiness.
Check shell commands for prohibited utilities in executable positions and suggest approved replacements. Trigger automatically before running or proposing commands that may use grep, find, cat, ls, or du.
Monitor GitHub Actions CI/CD workflows after git push. ALWAYS invoke this skill automatically immediately after any successful `git push` command — do not wait for the user to ask. Also invoke when the user says monitor CI, check if CI passed, watch the workflow, CI の結果を確認, ワークフローを監視, or provides a GitHub Actions URL. This skill is essential after pushing code, Claude must verify CI results proactively so the user never has to manually check GitHub.