The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content such as launch plans, claims, and assets, which creates a surface for indirect instructions to influence agent behavior.
Ingestion points: User-provided evidence passed as arguments and data retrieved via the WebFetch tool.
Boundary markers: No explicit prompt boundaries or delimiters are defined to isolate untrusted content from the agent's primary instructions.
Capability inventory: The skill has the ability to execute local Python scripts and write files to the memory/audits/ directory.
Sanitization: The skill utilizes a validation script (validate-audit-artifact.py) prior to persisting data, which mitigates but does not eliminate injection risks.
[COMMAND_EXECUTION]: The skill relies on shell command execution for environment configuration and task performance.
Evidence: Resolves the repository root using git rev-parse --show-toplevel to set the AARON_SKILLS_ROOT variable.
Evidence: Invokes scoring logic by executing python3 "$AARON_SKILLS_ROOT/scripts/rubric-score.py".
[DYNAMIC_EXECUTION]: The skill executes local Python scripts at runtime to perform its primary auditing functions.
Evidence: The instructions mandate the execution of rubric-score.py for scoring profiles and validate-audit-artifact.py for ensuring report integrity.

launch-readiness-auditor