Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/zhiganov/skill-driftnpx agentmods add commands/zhiganov/skill-drift/check-skillsWrote 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/commands/zhiganov/skill-drift/check-skills)<a href="https://agentmods.dev/commands/zhiganov/skill-drift/check-skills"><img src="https://agentmods.dev/badge/commands/zhiganov/skill-drift/check-skills/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/commands/zhiganov/skill-drift/check-skills"><img src="https://agentmods.dev/badge/commands/zhiganov/skill-drift/check-skills.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 | $0.00018 | $0.02481 |
| Opus 5 | $0.00009 | $0.01241 |
| Sonnet 5 | $0.00004 | $0.00496 |
| Haiku 4.5 | $0.00002 | $0.00248 |
Grade B, and why
check-skills scanned grade B with 1 finding 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 2d 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
Installing it on a new machine (machine-local — `~/.claude/settings.json` is not synced): How it starts
The opening of the file, as written. The whole thing — 160 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Check skills
Skills have no update banner, unlike plugins. This is the substitute: one read-only pass over every install mechanism on this machine, answering what is active, where did it come from, and is any of it behind?
Run it
node scripts/check-skills.mjs $ARGUMENTS
| Flag | Effect |
|---|---|
| (none) | Full report, refreshes the cache |
--json |
Machine-readable output |
--no-network |
Skip every remote call — everything becomes "undeterminable" rather than falsely reading as current |
--banner |
Session-start hook mode. Cache only, no network. Emits a JSON object or nothing — not for humans. |
--refresh |
Run the check, write the cache, print nothing |
Then show the user the report as-is. It is already formatted for a terminal — do not re-summarise it into a table of your own.
The session-start banner
A SessionStart hook runs --banner, which reads only the cache — no network, no git, no npm,
~75ms. If the cache is older than roots.cacheTtlHours (default 24) it spawns a detached refresh
whose result appears next session. Session start never waits on ~25 GitHub calls.
It emits one JSON object, or nothing. It fires only on states that are actionable now:
something behind upstream, something in the repo not linked on this machine, or the checkout
behind origin. Provenance gaps (undeterminable, unattributed) are deliberately excluded — they
are real but near-permanent, and a banner that fires on a healthy machine every session is one you
learn to scroll past.
It must be JSON, not plain text. A SessionStart hook's plain stdout is added to Claude's
context, not shown to the user — that is documented behaviour for SessionStart,
UserPromptSubmit and UserPromptExpansion, and it is the opposite of what a banner needs. The
only field that surfaces text to the person at the keyboard is systemMessage. The first version
of this shipped as plain text and was therefore invisible to its entire audience.
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.
- 2d ago Changed · +15 lines 4ee7ed600b87
- 9d ago First seen · 145 lines · 18 tokens per session scan B 3c71fac5402f
check-skills is a command published in the GitHub repository zhiganov/skill-drift (2 stars, last pushed 4d ago), licensed MIT. It adds 18 tokens to every session and 2,481 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
review-drift
Walk the VibeFrame drift inbox and resolve each finding (fix / accept / dismiss).
plan-feature
Plan and implement a feature through the VibeFrame model (context → declare → implement → link → validate).
reconcile
Explain unmapped code changes and fold them into the VibeFrame model.
ui-evidence
Discover, configure, and run ui-evidence for before/after UI review in this repo.
checklist
Generate a custom checklist for the current feature based on user requirements.
clarify
Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec.