The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a writing contract for reports intended to be parsed by a readVerdicts() function, creating a potential surface for schema confusion or indirect prompt injection.
Ingestion points: Agents utilizing these templates (such as reviewers and architects) ingest content from external sources including codebases and commit metadata (e.g., SKILL.md placeholders for file paths and commit SHAs).
Boundary markers: The templates do not define clear delimiters or instructions for the agent to ignore or escape instructions that might be embedded in the data being reviewed.
Capability inventory: The skill is scoped to Read and Write tools, allowing agents to process data and generate reports within the docs/ and .great_cto/verdicts/ paths.
Sanitization: The instructions lack guidance on validating or escaping untrusted input before it is included in findings or the terminal VERDICT line, which is used for machine-based decision making.

prose-style