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 Malo-T/claude-toolbelt --skill audit-contextgit clone --depth 1 https://github.com/Malo-T/claude-toolbeltWrote 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/malo-t/claude-toolbelt/audit-context)<a href="https://agentmods.dev/skills/malo-t/claude-toolbelt/audit-context"><img src="https://agentmods.dev/badge/skills/malo-t/claude-toolbelt/audit-context/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/malo-t/claude-toolbelt/audit-context"><img src="https://agentmods.dev/badge/skills/malo-t/claude-toolbelt/audit-context.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 | $0.00218 | $0.06093 |
| Opus 5 | $0.00109 | $0.03047 |
| Sonnet 5 | $0.00044 | $0.01219 |
| Haiku 4.5 | $0.00022 | $0.00609 |
Grade A, and why
audit-context scanned grade A 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 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.
Reads agent configuration directorieslowAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
`grep -o '"CLAUDE_CODE_GLOB_NO_IGNORE"[^,}]*' .claude/settings.json .claude/settings.local.json Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 367 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Audit Context
Read-only. Measures four sources of context bloat on disk, then weighs them against what past
sessions actually injected, and ranks the levers that address them.
No file is written and no setting is changed. Fixes to the repository's own reach live in
ignore-setup; fixes to the Claude Code setup around it — CLAUDE.md bulk, unused skills, plugins
and MCP servers — live in the built-in /doctor, not here (see ../../ROADMAP.md).
/clean-context:audit-context
This skill does not fix anything
If the ask mid-run turns into "et maintenant exclus-les", report what the measurement found and
hand off to ignore-setup; do not start editing. That boundary is the only thing separating the two
skills. If it turns into "désactive ce serveur" or "allège mon CLAUDE.md", hand off outside this
plugin instead: point at /doctor, which acts on both and writes the settings this skill only reads.
Step 0. State of play
- Project root:
git rev-parse --show-toplevel 2>/dev/null || pwd. No.gitis not an error: fall back topwdand say so in the report. Every measurement below still works, except thatrg's.gitignoresupport is git-repo-gated (Step 2 below adds--no-require-gitfor this case). - Detect whether
ignore-setupalready ran here:grep -o '"CLAUDE_CODE_GLOB_NO_IGNORE"[^,}]*' .claude/settings.json .claude/settings.local.json 2>/dev/null. If it is set tofalse, Glob already obeys ignore files this session, so label Step 2's two numbers accordingly instead of assuming the raw count is "today". - Detect whether CLAUDE.md loading is disabled entirely:
env | grep CLAUDE_CODE_DISABLE_CLAUDE_MDS. If set, Step 4's byte totals are moot. Say the budget is already zero and why, and skip straight to that row in the report.
Step 0bis. Tokenizer availability
Bytes are a proxy that misleads on minified assets and CJK text, and there is no local Claude
tokenizer to fall back on: the only exact count is the Messages API's count_tokens endpoint,
which needs network access and credentials and is too heavy for a diagnostic that should run
anywhere. So: use a real tokenizer opportunistically if one is already reachable, never install
one silently.
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 · 367 lines · 218 tokens per session scan A 351803f7e4e6
audit-context is a skill published in the GitHub repository Malo-T/claude-toolbelt (2 stars, last pushed 20d ago), licensed MIT. It adds 218 tokens to every session and 6,093 once invoked, about $0.0011 per session on Opus 5. A static security scan graded it A 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 skills, from other repositories
reflect
Per-project self-improvement - reads the .harness ledger and feedback memories, then proposes gated rule/threshold/ADR changes so the project stops repeating mistakes. Run periodically.
remember
Review auto-memory entries and propose promotions to CLAUDE.md, CLAUDE.local.md, or shared memory. Also detects outdated, conflicting, and duplicate entries across memory layers.
magic-compact
Compact this Claude Code session.
dream
Memory consolidation - review, merge, prune, and index memory files. Run periodically to keep memories organized and up-to-date.
cold-outreach
Research a prospect, evaluate fit, draft a personalized cold email, preview it to yourself, send it via Resend, and track it in a bounded ledger. Use when the user asks to "research a company/prospect", "check out this lead", "draft a cold email", "send cold outreach", "outreach to this business", "deep dive on this…
email-broadcasts
Draft, test, and send marketing / product-update email broadcasts to an audience via Resend. Use when the user asks to "send a broadcast", "email blast", "newsletter", "product update email", "announce something to users", "marketing email", "review request", "how many contacts", "email history", or "what have we…