The Agent Skills Directory

[SAFE]: The skill operates entirely within the local filesystem to perform its stated purpose of auditing credential hygiene. It identifies secret identifiers (keys) rather than their values and does not attempt to exfiltrate any data.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface as it reads files from other skills which are externally provided.
Ingestion points: The audit.py script recursively reads all non-binary files within the SKILL_SEARCH_PATHS (e.g., ~/.openclaw/extensions) to scan for credential patterns.
Boundary markers: The skill does not implement specific boundary markers or "ignore" instructions when processing these files, though it uses regular expressions to isolate specific keywords.
Capability inventory: The audit.py script contains no high-risk capabilities such as network operations, subprocess execution, or dynamic code evaluation. Its actions are limited to reading files and writing a localized state file (state.yaml).
Sanitization: The script uses read_text(errors="ignore") and matches against specific CREDENTIAL_PATTERNS to extract identifiers, which serves as a basic filter against arbitrary instruction execution.

secrets-hygiene