The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves parsing and interpreting local development logs (JSONL files). These logs may contain untrusted or malicious instructions from previous sessions or external data sources that could influence the agent's behavior during the audit.
Ingestion points: Phase 2 in SKILL.md directs the agent to use probe_jsonl_schema.py to ingest and analyze files from discovered project roots.
Boundary markers: There are no explicit delimiters or structural markers used to isolate untrusted log content from the agent's core instructional context when generating interpretations or recommendations.
Capability inventory: The skill has the ability to read local files, execute bundled Python scripts, and perform filesystem write operations within the designated output directory.
Sanitization: The skill implements manual redaction guidelines in the 'Security boundary' section and utilizes a regex-based secret scanner (finalize_audit.py) to detect and block the inclusion of credentials in the final report.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and Python scripts using arguments and paths discovered from the local environment.
Evidence: SKILL.md contains several execution patterns, such as python3 "$OUT_DIR/scripts/probe_jsonl_schema.py" <discovered-roots...>, where the agent passes dynamically resolved filesystem paths directly into shell commands.

dev-environment-roi-audit