The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data (project documents, PRDs, analytics exports, and websites) in Step 1 of the workflow. This data is interpolated into the agent's context to generate the business plan outline, presenting a surface for indirect prompt injection.
Ingestion points: Step 1 in SKILL.md identifies repository files, URLs, and analytics exports as input sources.
Boundary markers: The instructions lack explicit delimitation (e.g., XML tags or clear separators) to isolate external content from the skill's operational instructions.
Capability inventory: The skill possesses file-system write capabilities (via the workspace initialization script) and the ability to process multiple data sources.
Sanitization: While the initialization script uses slugify for file paths, there is no evidence of sanitization for the textual content of ingested documents.
[COMMAND_EXECUTION]: The skill executes a local Python script scripts/init_bp.py during its mandatory workflow to set up the workspace.
Evidence: The workflow in SKILL.md invokes python3 "$SKILL_DIR/scripts/init_bp.py" --name "Project Name" --stage seed --output ./business-plan.
Analysis: The script is bundled with the skill and performs path resolution and template rendering. It includes safety checks (ensure_safe_destination) to prevent writing to sensitive directories like the system root or the user's home directory.

lov-bp-outline