fires when
Use before committing a NEW or EDITED skill, its references, its bundled scripts, or its eval fixtures to this marketplace — review the change for anything UNSAFE to the person who installs and runs it: prompt-injection text, data exfiltration, an embedded secret, a dangerous/destructive command, obfuscated payloads, an unexpected external fetch, credential-store access, or privilege escalation. The consumer-safety half of the gate; the outward-leak half is cerberus:leak-check. Also runs when the CERBERUS hook flags a skill/script/eval edit.
best for
Any marketplace where an installed skill runs on someone else's machine and must be certified safe, not just leak-free.
measured · trigger
best score
7/10
description-triggering measured 2026-09-08
measured · quality
Not measured yet. A quality number needs its own eval: 15 or more tasks, each run with the skill and without it.
security scan · 8/8
✓prompt-injectionbody addresses the human author; no directive aimed at a reading agent; the 2026-10-19 edit only renames the quality tool in prose; no new instruction to the agent beyond pointing at save-result.mjs --quality
✓data-exfiltrationno code that sends repo/user/context data anywhere; the edit names a local repo script that writes evals/result.json; nothing leaves the machine
✓secret-detectionno key, token, or credential embedded; the edit adds no key, token or credential
✓dangerous-commandsdestructive strings appear only as prose examples of what to look for; none executed; the edit adds no shell command beyond a mention of a local script that writes one JSON file
✓obfuscationplain markdown throughout; nothing encoded or unreadable-on-purpose; plain prose
✓external-fetchesno download-and-run or reach to an external host; the edit contacts no host
✓credential-accessno read of ~/.ssh, ~/.aws, .env, keychain, cookies, or a token store; the edit reads no credential store
✓privilege-escalationno sudo/setuid/persistence; no write outside the working tree; the edit adds no sudo, cron, profile or launch agent
scanned 2026-10-19 · cerberus:security-scan
side effects & deps
fileshook
needsnothing
what’s inside
SKILL.md7.0 KB
metadata.yaml602 B
evals/result.json5.2 KB
evals/rubric.json2.3 KB
references/checklist.md3.9 KB