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.
npx skills add brandonsimpson/devils-advocate --skill critiquegit clone --depth 1 https://github.com/brandonsimpson/devils-advocateWrote 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.
[](https://agentmods.dev/skills/brandonsimpson/devils-advocate/critique)<a href="https://agentmods.dev/skills/brandonsimpson/devils-advocate/critique"><img src="https://agentmods.dev/badge/skills/brandonsimpson/devils-advocate/critique.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00013 | $0.03383 |
| Opus 5 | $0.00006 | $0.01691 |
| Sonnet 5 | $0.00003 | $0.00677 |
| Haiku 4.5 | $0.00001 | $0.00338 |
Grade A, and why
critique 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 7d 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.
How it starts
The opening of the file, as written. The whole thing — 290 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Devil's Advocate Critique
You are running an adversarial binary critique. Every criterion either passes or fails — no percentage scores, no wiggle room. You must be your own harshest critic.
Target Detection
Determine whether you are critiquing code or a plan based on conversation context:
- If the user provides a path to a plan/design document, or you just wrote one → plan critique (22 criteria)
- If you just wrote code, or the user asks you to critique code changes → code critique (20 criteria)
- If unclear, ask the user
Scope-Bounded Critique
Critique ONLY what was requested. Do not penalize for out-of-scope features. If a criterion doesn't apply to the target (e.g., no-injection for a project with no user input), mark it PASS with a note — don't skip it.
Process
Step 0: Independence Gate
If you wrote or contributed to the target being critiqued (same conversation), you MUST dispatch an independent subagent. Same-context critique has author bias — the author fills in gaps mentally and rationalizes decisions. Independent critique only sees the artifact and codebase.
Dispatch pattern: Before dispatching, replace TARGET_PATH with the actual file path or description of changes, and expand the criteria placeholder with the appropriate criteria block from Step 4 (code or plan).
Agent({
description: "Independent DA critique",
prompt: `You are a devil's advocate reviewer. Perform an adversarial binary critique of: TARGET_PATH
Read CLAUDE.md first for project conventions. Then read the target. Then verify all claims against the actual codebase using Read, Grep, Glob, and Bash — run the project's own test/build/typecheck commands (discover them from the package manifest, Makefile, or CLAUDE.md; prefer quiet/summary flags where available).
CRITERIA_BLOCK
For each criterion: think step by step before marking it PASS or FAIL. Then: PASS with brief evidence, or FAIL with file:line and a Fix: suggestion.
Then write the logs (paths relative to the project root):
1. Read .devils-advocate/session.md if it exists. Find the highest check number and use the next one as N. Get the commit SHA via git rev-parse --short HEAD and the timestamp via date +"%Y-%m-%d %H:%M".
2. Append this entry to .devils-advocate/session.md, preserving all existing content (create the file if missing):
## Check #N — Critique | YYYY-MM-DD HH:MM | <git-sha>
- **Result:** X/Y PASS
- **Failing:** [comma-separated failing criteria, or "none"]
- **Summary:** [1-2 sentence summary]
3. Write the full formatted critique output to .devils-advocate/logs/check-N-critique-YYYY-MM-DD-HHMM.md, creating the logs/ directory if needed.
4. Run: touch .devils-advocate/.commit-reviewed
Return the full formatted critique as your final message.`
})
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.
- 7d ago First seen · 290 lines · 13 tokens per session scan A 950c7180c84e
critique is a skill published in the GitHub repository brandonsimpson/devils-advocate (22 stars, last pushed 2mo ago), licensed MIT. It adds 13 tokens to every session and 3,383 once invoked, about $0.0001 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 2026-08-30.
Other skills, from other repositories
review-responder
Use when receiving feedback on code you implemented, responding to reviewer comments, deciding which review suggestions to implement, or pushing back on incorrect review feedback.
rubber-ducky
Use when you've planned a non-trivial change and are about to implement it, finished a complex or multi-file piece of work, just wrote tests, or are stuck on repeated failures — and any time the user says "rubber duck this", "rubber ducky", "get a second opinion", "sanity-check my plan", "poke holes in this", "what am…
pr-sweep
Use when you want to sweep all open pull requests across all repos, triage their status, run code reviews on unreviewed PRs, merge what's ready, fix quick blockers, and produce a full status report. Trigger when the user says "check my PRs", "close out open PRs", "what's the status of my PRs", "sweep my PRs", "PR…
review
Use when user invokes /review to review code changes. Also triggers on "review my code", "check this PR", "look at my changes". Accepts a PR number (e.g. /review 123), a path filter (e.g. /review src/auth/), or no argument (reviews current branch vs base). Produces structured per-file review with severity labels…
open-pr
Use when wrapping up a development task and getting a PR ready — runs local tests, creates a PR (if one doesn't exist) with a structured description template, conducts a code review via the review skill, and checks CI status with quick fixes. Trigger when the user says they're done with a feature, want to open a PR…
session-review
End-of-session adversarial review loop. Assemble the session's work into a role-assigned, self-contained brief, then run independent reviewers in parallel — an isolated code-reader (the idea-validator agent) that reads the ACTUAL files and web-checks technology currency, plus an external-family model if you have one …