The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external project documentation to verify implementation alignment. While this creates a potential surface for indirect injection if project files contain malicious instructions, it is a primary function of the skill's drift alignment logic.
Ingestion points: Reads from proposal.md, design.md, visual.md, and memory/ARCHITECTURE.md to compare against implementation.
Boundary markers: None explicitly defined in the instructions for these file reads.
Capability inventory: The skill possesses file-writing capabilities (modifying CLAUDE.md, memory/FAILURES.md, and source code) and shell execution capabilities (running tests and linters).
Sanitization: No explicit sanitization or escaping of the ingested file content is described.
[COMMAND_EXECUTION]: The skill requires the agent to execute various shell commands to generate evidence of testing, including linting, type checking, unit tests, and end-to-end (E2E) suites. This is consistent with the skill's stated purpose of providing physical evidence of code correctness.

fancy-harden