The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze potentially untrusted data from session history, subagent reports, and code diffs to reconstruct decision ledgers.
Ingestion points: Untrusted data enters the context through session history, subagent reports, transcripts, and diffs during the 'Elicit and trace back' phase.
Boundary markers: The instructions do not prescribe the use of delimiters or 'ignore' instructions for the data being analyzed.
Capability inventory: The skill permits file system writes (decisions.md), execution of existing project tests, and the creation/execution of transient probes.
Sanitization: No specific sanitization or filtering of external content is mentioned; the agent is relied upon to triage based on evidence.
[DYNAMIC_EXECUTION]: The workflow allows for the creation and execution of code probes to gather evidence during the audit.
Evidence: The 'Rules' section states that writing transient probes to verify code properties is permitted, which involves runtime script generation and execution based on potentially attacker-influenced code patterns.
[COMMAND_EXECUTION]: The agent is explicitly instructed to run existing tests within the project environment to gather evidence for the audit verdicts.

audit-decisions