القائمة
Use when asked to watch, babysit, audit, review, compare, or fix another agent's work from a session ID, Claude Code session/transcript, PR, branch, log, or run summary.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | agent-watchdog |
| description | Use when asked to watch, babysit, audit, review, compare, or fix another agent's work from a session ID, Claude Code session/transcript, PR, branch, log, or run summary. |
Watch another agent's work like a reviewer with a pager: wait for completion when needed, reconstruct the request, verify the evidence, and close the gap between what was asked and what actually happened.
Infer the mode from the user's wording:
If authority is unclear, default to audit-only and say what you would fix.
Build a compact contract before judging the work:
Treat the user's request as the source of truth, not the other agent's summary.
Inspect evidence, not vibes:
Classify each issue as:
When the user authorized repair:
Lead with the outcome. Keep the report short enough to scan:
Status
- Done, blocked, stale, or still running.
Requested
- What the user asked the watched agent to do.
Observed
- What the watched agent changed, claimed, and verified.
Gaps
- Missing behavior, bugs, weak verification, or scope drift.
Fixes made
- Files changed and validation run. Omit this section for audit-only work.
Remaining risk
- Anything still unverified or waiting on CI/review/deploy/human input.
Name exact files, commands, PRs, or thread IDs when they matter.
The autonomous-sdlc state machine. Loaded by every /sdlc loop iteration to decide what one unit of work the current state requires, how to verify it, and which transition to record. Trigger: an active `.sdlc/state.json` exists, or the /sdlc command invokes it. Not for ad-hoc use outside a loop.
Save agent-skills workflow feedback proactively. MUST trigger when the user corrects how a visual plan/recap, recap status block, orchestration, autonomy, docs-first, or review skill behaves ("recaps are too long", "always web-search the docs first", "don't stop to ask", "plans should lead with the wireframe") or confirms good output. Also trigger on explicit requests: "save this feedback", "show my feedback", "clear feedback", "agent-skills preferences", "consolidate feedback", "bake in my preferences". Feedback loaded automatically when running these skills.
Use always when operating as Claude Fable on codebase-heavy or token-heavy work: Fable orchestrates, architects, synthesizes, and judges while cheaper subagents do bounded research, coding, testing, and log reduction.
Use whenever running a high-cost frontier model on codebase-heavy or token-heavy work (apply automatically on such models): delegate research, coding, testing, and log reduction to cheaper subagents, reserving the expensive model for planning, tradeoffs, synthesis, and final review.
Use when asked to compare, cross-review, merge, judge, choose, or arbitrate competing plans from multiple agents such as Codex and Claude Code; when given two or more proposed plans, session IDs, transcripts, plan documents, PR descriptions, or pasted strategies.
Use when the user explicitly wants autonomous progress without routine clarification stops: "plow ahead", "do not stop", "use your best judgment", "keep going until done", "finish while I am away", "do not ask questions unless truly blocked", or similar. Convert ordinary ambiguity into stated assumptions, proceed through implementation and validation, stop only for true blockers, and end with a clear recap of decisions, changes, verification, and residual risk.