The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for adopting existing project materials (PRDs, specs, design docs) and mining external competitor reviews. This creates a surface for indirect prompt injection where untrusted data could influence the agent's behavior during the drafting phase.
Ingestion points: references/migration.md (Inventory of legacy documents) and references/research.md (Competitor review mining).
Boundary markers: The skill lacks explicit instructions for the agent to wrap or delimit external content to prevent instruction override.
Capability inventory: The skill utilizes file system write operations, web search, and local code execution capabilities.
Sanitization: No specific sanitization or filtering logic is prescribed for the ingested text.
[DYNAMIC_EXECUTION]: The skill encourages the use of runnable code (specifically economics.py or spreadsheets) within the research/ directory to maintain and update financial models. This involves the agent generating or re-running local scripts to compute business metrics.
[CREDENTIALS_UNSAFE]: The skill includes a dedicated section for defining a 'secrets policy' within stack.md. It follows safe practice by explicitly stating that only the 'location and policy' should be recorded, 'never a value', effectively preventing the hardcoding of sensitive credentials within the generated blueprint files.

blueprint