The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on shell execution of included Python scripts (scripts/init_bp.py, scripts/audit_bp.py) to perform workspace management and deterministic linting of business plan content. These operations are scoped to the skill's local environment and intended workspace.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because its core function involves processing untrusted external data sources, including project repositories, documentation, and web pages, to construct investment narratives. Malicious content within these inputs could potentially influence agent behavior.
Ingestion points: The lov-bp-outline module explicitly reads repository files, product docs, PRDs, websites, and customer notes (skills/bp-outline/SKILL.md).
Boundary markers: The workflow employs 'evidence gates' and mandates the use of an evidence-ledger.md to classify claims as facts, inferences, or assumptions, which serves to compartmentalize untrusted data.
Capability inventory: The skill can perform file system writes (workspace setup), execute local Python scripts, and access network resources for research purposes.
Sanitization: The audit_bp.py script provides deterministic checks for structural integrity and placeholder markers, ensuring generated content adheres to a predefined investor rubric rather than purely responding to input data.

lov-bp