The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill auditing process involves reading untrusted source code and passing it to an agent swarm. The resulting findings are interpolated into a Markdown report. While the report contains clear non-authoritative labels, it does not explicitly sanitize findings for prompt injection patterns, which could potentially be processed by other automated systems.
Ingestion points: Source code targets specified by the user via /woostack-audit <target>.
Boundary markers: The report generated in scripts/render-report.sh includes a header: 'Non-authoritative diagnostic evidence — report only.'
Capability inventory: The skill reads local files, writes reports to .woostack/audits/, and executes system binaries (git, jq, python3).
Sanitization: scripts/render-report.sh implements regex-based redaction for credentials, personal data, and local system paths.
[COMMAND_EXECUTION]: The skill uses shell scripts to orchestrate the audit process, executing git for repository enumeration and diffing, jq for manipulating JSON metadata, and python3 for logic processing.
[DYNAMIC_EXECUTION]: In scripts/load-audit-config.sh, the skill dynamically modifies the Python module search path using sys.path.insert to load a configuration helper from a relative sibling directory (../../woostack-review/scripts).

woostack-audit