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 agentmods add skills/grinv/mal-mcp/mutation-test-safetynpx skills add Grinv/mal-mcp --skill mutation-test-safetygit clone --depth 1 https://github.com/Grinv/mal-mcpWrote 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/grinv/mal-mcp/mutation-test-safety)<a href="https://agentmods.dev/skills/grinv/mal-mcp/mutation-test-safety"><img src="https://agentmods.dev/badge/skills/grinv/mal-mcp/mutation-test-safety.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 | $0.00051 | $0.00491 |
| Opus 5 | $0.00026 | $0.00246 |
| Sonnet 5 | $0.00010 | $0.00098 |
| Haiku 4.5 | $0.00005 | $0.00049 |
Grade A, and why
mutation-test-safety 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.
What it actually says
Mutation testing safety contract
Applies to any mutation tool (add/update/remove/delete/post/follow/favourite/ etc.) once the user has given explicit go-ahead to test it against a real account. For every mutation call:
- Capture the exact pre-state first (via the matching read tool) — not
just an assumption of what it probably is. A target with no existing
state has a clear pre-state too: "absent." This applies to every call
to a mutation tool, including one you intend only as a validation-only
probe — if the schema isn't
.strict(), a bogus/typo'd field is silently dropped rather than rejected, and any other real field in the same call still applies for real. - Make the smallest possible change that still exercises the behavior (e.g. one field, not a full rewrite). Never combine an unknown/invalid probe field with a real valid field in one call — if the probe field is ignored instead of rejected, the valid field mutates the account for real. Test unknown-param rejection with no other fields set (or on a read-only tool) instead.
- Verify the change landed by re-fetching via a read tool — a mutation's own echoed response is not always trustworthy (some tools have historically omitted fields they actually changed).
- Revert to the captured pre-state immediately, in the same turn, and verify the revert too. Don't batch several mutations and revert at the end — revert each one before moving to the next unrelated test.
- Never leave the target in a different state than you found it, even if a step errors partway through — check and clean up regardless.
- Never touch an uninvolved third party. Self-targeted mutations (a self-message, a self-created and immediately-deleted test post/ thread/comment) are fine; acting on a random other real user/account is not.
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.
- 6d ago First seen · 37 lines · 51 tokens per session scan A 83101f0a32ff
mutation-test-safety is a skill published in the GitHub repository Grinv/mal-mcp (2 stars, last pushed 12d ago), licensed MIT. It adds 51 tokens to every session and 491 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.
Other skills, from other repositories
live-audit
Audit anilist-mcp-server — build/test/lint gate, live MCP tool edge-case sweep (input validation, not-found paths, mutations with capture/revert), source-level code review, and docs/metadata consistency. Use when asked to test/audit the published or just-fixed anilist-mcp-server package, hunt for bugs/edge cases, or…
tool-description-check
Self-check a new or edited MCP tool description/field .describe() text before committing — verify every behavioral claim against live testing, check for contradictions with sibling tools, and score against Glama's Tool Definition Quality Score (TDQS) rubric. Use whenever a tool description or schema field description…
release
Cut a release of anilist-mcp-server — draft CHANGELOG entries, check docs/metadata consistency, then bump/tag/push. Use when asked to release, cut a version, or publish a new version of this package.
fixture-accuracy-check
Make sure a mocked-fetch test fixture mirrors AniList's real GraphQL response shape, not just whatever fields make the current code pass. Use before writing or changing a fixture in src/tests/.test.ts.
docs-consistency-check
Check README/manifest.json/server.json/CHANGELOG.md/AGENTS.md and docs/.md for drift against the actual registered tools and source. Use after adding, renaming, or removing a tool, or as part of a live-audit pass.
mutation-test-safety
The capture-state/smallest-change/verify/revert/verify-revert contract for live-testing a mutation tool against a real account. Use any time you're about to call a mutation tool live, not just during a full audit.