agent-watchdog

agent-watchdog is a skill for Claude Code from tslateman/duet. It costs 63 tokens per session (979 once invoked), scanned A, original, MIT.

A review tool for checking whether another coding agent actually completed the requested work, including any claimed tests or verification.

In plain words
What is it for?
Use it to watch a session or workflow, audit a transcript, pull request, branch, commit, or CI run, compare multiple agent runs, or make narrow fixes after an audit.
Why use it?
A finished-looking response or clean code does not prove that the original request was fulfilled. It exposes missing work, unsupported claims, and gaps between the request and the result.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: mentions Claude Code; mentions Codex.

Part of the duet plugin — 25 skills, 6 commands, 1 agent shipped together

Good fit Use it to watch a session or workflow, audit a transcript, pull…

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/tslateman/duet/agent-watchdog
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add tslateman/duet --skill agent-watchdog
Clone the repo
git clone --depth 1 https://github.com/tslateman/duet

Made for: Claude Code.

Or install duet, the plugin that ships this one along with the rest of its 25 skills, 6 commands, 1 agent.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for agent-watchdog

README.md
[![agentmods](https://agentmods.dev/badge/skills/tslateman/duet/agent-watchdog.svg)](https://agentmods.dev/skills/tslateman/duet/agent-watchdog)
Your own site
<a href="https://agentmods.dev/skills/tslateman/duet/agent-watchdog"><img src="https://agentmods.dev/badge/skills/tslateman/duet/agent-watchdog.svg" alt="Measured on agentmods" height="20"></a>
Per session 63 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 979 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00063 $0.00979
Opus 5 $0.00032 $0.00490
Sonnet 5 $0.00013 $0.00196
Haiku 4.5 $0.00006 $0.00098

Measured 6d ago against content hash a1712c9f5777, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

agent-watchdog scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/agent-watchdog/SKILL.md · 89 lines

How it starts

The opening of the file, as written. The whole thing — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agent Watchdog

Watch another agent's work like a reviewer with a pager: reconstruct the request, verify the evidence, and close the gap between what was asked and what actually happened. This audits the agent (did it do the task, did it really verify), not the code itself; pair it with /review or /code-review for the code.

Adapted from BuilderIO/skills (MIT).

Choose the mode

Infer the mode from the user's wording:

  • Watch only: monitor a session, PR, branch, CI run, or workflow until it reaches a terminal state. No edits.
  • Audit: read the prompt, transcript, diff, tests, CI, and final claims, then return a gap report. No edits.
  • Audit and fix: audit first, then make narrow fixes for clear gaps. No broad rewrites, branch moves, or speculative changes.
  • Compare: given multiple agents or runs, judge each against the same original request and reconcile the differences.

If authority is unclear, default to audit-only and say what you would fix.

Resolve the target

  1. Identify every artifact supplied: session or transcript path, thread URL, PR, branch, commit, CI run, issue, or pasted summary.
  2. Resolve it from the most direct source (local transcript, repo logs, gh for PR/CI), not a summary.
  3. If it is still running and you were asked to watch, poll at a reasonable interval until it is done, blocked, stale, or waiting on a human.
  4. If it cannot be resolved, ask for the missing identifier.

Reconstruct the contract

Build a compact contract before judging the work:

  • The original request and any later scope changes. This is the source of truth, NOT the agent's own summary.
  • Explicit constraints: branch rules, no-edit requests, versions, validation expectations, security or privacy limits.
  • Implied acceptance criteria: user-visible behavior, tests, CI, docs, deploys, review replies.
  • The agent's final claims and its "could not do" caveats.

Audit the evidence

Inspect evidence, not vibes:

  • Read the changed files and the relevant unchanged files around them.
  • Check git status and diff without reverting unrelated work.
  • Compare the commands the agent claimed to run against actual output.
  • Inspect failed or skipped tests, CI logs, and error traces; for UI work, prefer screenshots or a browser check over prose claims.
  • For PR work, verify unresolved threads and CI state from the source system.

Read the full file on GitHub · 89 lines

Files

What ships with it

1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 6d ago First seen · 89 lines · 0 tokens per session scan A a1712c9f5777

Subscribe to this mod's changes

agent-watchdog is a skill published in the GitHub repository tslateman/duet (1 stars, last pushed 10d ago), licensed MIT. It adds 63 tokens to every session and 979 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

self-audit

Self-audit this skills repo against CLAUDE.md invariants, the artifact contract, and README/CLAUDE.md/docs sync via three parallel read-only subagents. Local meta-skill — independent of the /task: pipeline.

SpaiR/task-pipeline · 52 tokens

gate-check

Final quality gate before reporting task completion (Gate 3). Fills confidence assessment, runs supplementary verification for <95% items, completes self-check checklist. Writes PASS/FAIL result to subtask.md, then calls autoworker:dispatch for routing.

phj128/autoworker · 55 tokens

verifying-task-output

Use after a dispatched task returns, to verify the output before marking it done.

tunahanaliozturk/atelier · 21 tokens

flow-next-resolve-pr

Resolve PR review feedback. Fetches unresolved threads, triages, fixes, replies and resolves via GraphQL. Use when asked to address review comments.

gmickel/flow-next · 36 tokens

critical-code-reviewer

Rigorously review code or pull requests for correctness, security, accessibility, maintainability, tests, and edge cases. Use when users request a critical code review, want a guided walkthrough of findings, need implementer-facing feedback, or want to prepare, create, or submit a GitHub pull request review.

posit-dev/skills · 67 tokens

brooks-sweep

Full-sweep mode: runs a unified analysis across all quality dimensions — code decay, architecture, tech debt, and test quality — then applies fixes directly to the codebase. Safe changes are auto-applied; risky changes are confirmed before execution. Drawing on twelve classic engineering books. Triggers when: user…

hyhmrright/brooks-lint · 178 tokens