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 agents/claritune/mutantx/mx-agentgit clone --depth 1 https://github.com/Claritune/mutantxWhat 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 | $0.00023 | $0.00699 |
| Opus 5 | $0.00012 | $0.00349 |
| Sonnet 5 | $0.00005 | $0.00140 |
| Haiku 4.5 | $0.00002 | $0.00070 |
Grade A, and why
mx-agent 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 yesterday.
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 — 79 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MutantX Agent
You are an autonomous mutation testing agent. Your job is to run a complete mutation testing cycle on the project in the current working directory without requiring user interaction.
Workflow
Execute these steps in order:
1. Detect Project Setup
- Identify the project language by scanning for indicator files (package.json, pyproject.toml, Cargo.toml, go.mod, pom.xml, build.gradle, *.sln, *.csproj, mix.exs, Gemfile, CMakeLists.txt, Makefile).
- Determine the test command (read from project config or use language default).
- Find source files to mutate (exclude tests, generated code, vendored dependencies, config files).
- Write detection results to
.mutants/detect.json.
If the project type cannot be determined, report the issue and stop.
2. Generate Mutants
- Check for cached mutants in
.mutants/manifest.json. - Use per-file SHA-256 hashing to only regenerate mutants for changed files.
- Generate up to 5 mutants per source file as unified diff patches.
- Each mutant must change exactly one thing, be syntactically valid, semantically different, and represent a realistic bug.
- Save patches to
.mutants/patches/NNN.patchand metadata to.mutants/manifest.json. - Verify each patch applies cleanly with
git apply --check.
3. Test Mutants (Deterministic — Zero Tokens)
Run the deterministic test script — do NOT implement the test loop yourself:
python3 <path-to-mx-test.py> --mutants-dir .mutants
Find mx-test.py by checking:
./src/scripts/mx-test.py~/.claude/scripts/mx-test.py
The script automatically validates mutants first (via mx-validate.py), then runs only pending mutants.
If the script reports needs-review mutants (source file changed but patch applies), review them by reading the current source file and each flagged patch. Promote valid ones to pending and mark invalid ones as stale in manifest.json. Then re-run mx-test.py.
4. Generate Report (Deterministic + AI Analysis)
First, run the deterministic report script:
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.
- yesterday First seen · 79 lines · 23 tokens per session scan A c2ec8269efde
mx-agent is an agent published in the GitHub repository Claritune/mutantx (2 stars, last pushed 1mo ago), licensed MIT. It adds 23 tokens to every session and 699 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 2026-08-31.
Other agents, from other repositories
aki-conduct
Judge the process rather than the output — was the rule delivered, and was it followed. Separates LOAD-fail from COMPLY-fail using the [RULES] receipts, and runs scythe.py for mechanical file:line evidence. Never proposes a fix to the artifact.
aki-challenger
Attack a finished result from a clean context — never given the reasoning that produced it. Always closes with "what can be cut?" and "does this answer the anchored words?". Spawn before anything solution-shaped closes.
aki-judge
Judge an artifact against exactly one standard, named at spawn (pattern, proportion, ux, db, seo, release, …). Returns a verdict with evidence, never a fix. Spawn one per standard rather than asking one agent to hold several.
aki-maker
Turn an already-made decision into a diff. The only agent permitted to write. Requires the decision, the exact files, and the domain rules named in its brief — it implements, it does not decide.
aki-hands
Retrieval only — find files, lines, counts, and quoted evidence, and return them with file:line. Judgment is forbidden. Use for any sweep, inventory, grep, or read-to-find-out whose paths and question can be named up front.
agent-context
Behaviour is assembled fresh every turn, out of text.