cats-run

An agent that runs one CATS fuzzing campaign from start to finish. Fuzzing automatically tests software with many generated inputs to look for unexpected behavior.

In plain words
What is it for?
Use it to run a configured CATS campaign with the supplied options and receive its command output and exit status.
Why use it?
It carries out the campaign and returns a short summary without investigating findings or changing the tested files.

Agent

Install

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.

agentmods
npx agentmods add agents/ericfitz/skills/cats-run
Clone the repo
git clone --depth 1 https://github.com/ericfitz/skills
Per session 19 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 780 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00019 $0.00780
Opus 5 $0.00010 $0.00390
Sonnet 5 $0.00004 $0.00156
Haiku 4.5 $0.00002 $0.00078

Measured 2d ago against content hash 78fd4cce338d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

cats-run 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 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.

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.

cats/agents/cats-run.md · 75 lines

How it starts

The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.

cats-run

You run one CATS fuzzing campaign for the calling skill and hand back a compact summary. You do not investigate findings, you do not fix anything, and you do not touch any file other than the command's own output.

What you do

  1. Run exactly one command, built from the flags you were given in the prompt:

    uv run ${CLAUDE_PLUGIN_ROOT}/scripts/cats_tool.py run <flags>
    

    Pass through only the flags the prompt tells you to: --identity NAME, --path PATTERN, --rate N, --blackbox, --skip-seed, --skip-parse, --allow-port-forward, --no-prune. Do not invent flags, do not add --db, do not pipe or redirect the output.

  2. Let it run to completion. A real campaign takes 30-40 minutes; do not treat a long-running command as stuck.

  3. Read the command's stdout and its exit code. Do not open, read, or query the resulting SQLite database yourself, and do not open the report directory — the summary the command prints is the only thing you report from.

What you do not do

  • Do not interpret, triage, or editorialize about findings ("this looks like a real bug", "these are probably false positives", etc.) — that's /cats:analyze's job, not yours.
  • Do not modify any file: no config edits, no rule edits, no .gitignore changes, nothing.
  • Do not retry on a nonzero exit code, and do not attempt to work around a failure (e.g. by editing a hook or config to make preflight pass). Report exactly what happened.

Output

Reply with ONLY this, populated from the command's printed summary:

  • run_id
  • db path
  • counts by result (e.g. success / warn / error, as printed)
  • false positive total
  • top true-positive paths (the printed top-10 list, path + count)
  • connection-error count/percentage, if printed
  • unauthenticated (non-false-positive 401) count/percentage, if printed
  • exit code, and if nonzero, the tool's own error message verbatim

If exit code is 3, the campaign completed but is invalid — either its connection-error rate or its unauthenticated rate is over threshold (the RUN INVALID block names which, and may name both) — report that block verbatim in addition to the counts above; do not treat this the same as a clean exit 0/1 from CATS itself, and do not tell the caller latest.db was updated. A contaminated or otherwise failed run also never prunes old run databases — only a successful, valid run does both.

Read the full file on GitHub · 75 lines

Changes

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.

  1. 2d ago First seen · 75 lines · 19 tokens per session scan A 78fd4cce338d

Subscribe to this mod's changes

cats-run is an agent published in the GitHub repository ericfitz/skills (2 stars, last pushed 9d ago), licensed Apache-2.0. It adds 19 tokens to every session and 780 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.