# Skill capability matrix — sample

Record what the skill can cause, not what its description says it does.

| Capability | Evidence to inspect | Default concern |
|---|---|---|
| Shell or process execution | commands, subprocess calls, script hooks | arbitrary host effects |
| Filesystem read | paths, globs, traversal, symlink handling | private source or config exposure |
| Filesystem write/delete | target resolution, overwrite and rollback rules | irreversible repository change |
| Network access | domains, methods, payloads, redirects | source or secret exfiltration |
| Secret/environment access | variable names, config readers, credential helpers | credential disclosure or misuse |
| Dynamic code loading | eval, dynamic import, decoded payloads | review bypass |
| Package installation | package manager commands, version pins, lifecycle scripts | supply-chain execution |
| External mutation | posting, billing, deployment, account or data changes | authority escalation |

For every present capability, record its trigger, target boundary, required
approval, data sent or changed, failure mode, and rollback path.

Sample from Skill Vetting Pack v1.0.0. The paid archive adds the local CLI,
decision record, change comparison, re-vetting triggers, playbooks, and tests.