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 niels-emmer/myace --skill drift-checkgit clone --depth 1 https://github.com/niels-emmer/myaceWrote 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/niels-emmer/myace/drift-check)<a href="https://agentmods.dev/skills/niels-emmer/myace/drift-check"><img src="https://agentmods.dev/badge/skills/niels-emmer/myace/drift-check/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/niels-emmer/myace/drift-check"><img src="https://agentmods.dev/badge/skills/niels-emmer/myace/drift-check.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00038 | $0.00728 |
| Opus 5 | $0.00019 | $0.00364 |
| Sonnet 5 | $0.00008 | $0.00146 |
| Haiku 4.5 | $0.00004 | $0.00073 |
Grade A, and why
Agent Instruction Drift Check 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 10d 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 — 40 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Purpose
Agent-facing instruction files (AGENTS.md, CLAUDE.md, or similar) are read by coding agents at the start of a task, not exercised by tests or compilers — nothing forces them to stay correct when the code changes underneath them. A stale instruction file actively misleads whichever agent reads it next: it can point at a command that no longer exists, describe a directory layout that moved, or state a rule the codebase no longer follows. Use this skill periodically, before a release, or whenever you suspect an instruction doc has gone stale.
When to use it
- Before cutting a release, as part of a documentation pass.
- After a significant refactor, migration, or dependency upgrade that could have invalidated commands or paths described in the doc.
- When an agent visibly acts on a rule that turns out to be wrong or outdated — that's a signal to check the whole doc, not just the one rule.
- On a regular cadence (e.g. monthly) for actively-developed projects, even with no specific trigger.
Procedure
- Inventory the doc's claims. Read through the instruction file section by section and list every concrete, checkable claim: commands, file paths, directory structures, described behaviors, architectural statements, named tools or dependencies.
- Verify each claim against the real thing, not against memory or the doc's own earlier version:
- Commands: actually run them (or confirm they exist in package.json/Makefile/pyproject.toml/etc.) with the flags the doc shows.
- File paths: confirm the file or directory still exists at that path.
- Described behavior: read the current implementation and confirm it still does what's described.
- Named tools/dependencies: confirm they're still in use (check the dependency manifest) rather than replaced or removed.
- Classify each finding:
- Broken — command/path no longer exists, will fail if followed.
- Stale — still technically works but describes old behavior, defaults, or structure.
- Unclear — doesn't contradict the code but wouldn't be enough for a reader who wasn't there to act on correctly.
- Accurate — confirmed correct, no action needed.
- Propose a specific fix for each Broken or Stale finding — the corrected command, path, or description — rather than just flagging that something's wrong.
- Check for cross-doc duplication while you're in there. If the same fact also appears in a human-facing doc (README, docs/), confirm both were checked and both get the same fix — don't correct one copy and leave the other stale.
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.
- 10d ago First seen · 40 lines · 38 tokens per session scan A 150b58cfd299
Agent Instruction Drift Check is a skill published in the GitHub repository niels-emmer/myace (1 stars, last pushed 3d ago), licensed MIT. It adds 38 tokens to every session and 728 once invoked, about $0.0002 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.
Other skills, from other repositories
loom-code-review
Comprehensive code review covering correctness, maintainability, performance, security, and best practices.
pressure
Pressure-test a loom plan and write a review for the plan's author. Trigger when the user invokes $pressure followed by a plan path. The argument after $pressure is the plan file path.
cross-review
The writer cannot be the final judge of the work.
code-review
Perform thorough code reviews focusing on correctness, security, and maintainability.
review
Multi-model code review of uncommitted changes. Spawns parallel reviewers across providers (Anthropic reuse/quality/efficiency + Codex grounded review) so blind spots in one model are caught by another. Use after writing or modifying code, before opening a PR, or when /simplify alone feels insufficient. Codex…
slop-check
LLM-judge slop & erosion scan of a diff, mirroring SlopCodeBench. Scores code on Erosion (verbosity, dead branches, redundant structure accumulated under iterative change) and Verbosity (unnecessary complexity), then reports per-category findings weighted toward code that EXTENDS existing modules. Use after…