The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local repository scripts using the Bash tool, including scripts/check-skill-integrity.py, scripts/validate-findings.py, and ./scripts/backtest.sh to perform mechanical verification and consistency audits.
[DYNAMIC_EXECUTION]: As part of the 'mechanical battery' verification process, the skill instructs the agent to write and execute scripts (e.g., population computations for estimating equations) to numerically verify mathematical identities.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted artifacts such as research papers and codebases, which provides a surface for indirect prompt injection.
Ingestion points: Content is ingested through the Read, Grep, and Glob tools for analysis by the auditor agents.
Boundary markers: The instructions emphasize quoting source text and grounding findings in the artifact, but do not provide specific prompt engineering delimiters to isolate ingested data from instructions.
Capability inventory: The agent possesses powerful capabilities including Bash for command execution and Write/Edit for file system modifications.
Sanitization: The workflow incorporates an independent 'Judge' agent to adjudicate findings, acting as a secondary verification layer before any findings are confirmed.

deep-audit