The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts, specifically rubric-score.py and validate-audit-artifact.py, to perform its core auditing and validation tasks. It also uses git rev-parse to resolve the root directory of the installation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external data such as launch assets, claims, and analytics evidence. While this represents an attack surface for indirect instructions, the skill mitigates this by using a deterministic scorer and an artifact validator to handle the data.
Ingestion points: Launch evidence files, rendered assets, narrative claims, and official platform rules mentioned in SKILL.md.
Boundary markers: The instructions enforce a structured RAMP framework with specific lifecycle profiles (preflight, execution, outcome) and item IDs to constrain the agent's interpretation.
Capability inventory: Includes execution of local Python scripts, shell command execution for environment resolution, writing to a protected audit directory, and fetching external data for verification.
Sanitization: Data is processed through a dedicated Python-based scoring engine and the final artifact is validated against a formal schema before persistence.

launch-readiness-auditor