The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a process to ingest project data to tailor design outputs, which introduces a potential attack surface.
Ingestion points: As described in SKILL.md and references/teach-impeccable.md, the agent is instructed to read README.md, package.json, .impeccable.md, .cursorrules, and various source files.
Boundary markers: There are no specific instructions or delimiters provided to ensure the agent ignores potentially malicious instructions embedded in the external files it reads.
Capability inventory: The skill has access to Bash, Read, and Write tools, allowing it to modify the file system and execute shell commands based on the context it gathers.
Sanitization: No validation or sanitization process is defined for the content ingested from the project's files.
[COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool for project discovery.
Evidence: In references/extract.md and references/normalize.md, the agent is told to "grep" the codebase for specific terms such as "design system" and "ui" to understand existing patterns.
[PERSISTENCE]: The skill uses repository files to maintain state across sessions.
Evidence: references/teach-impeccable.md requires the agent to save the gathered design context into .impeccable.md and potentially .cursorrules in the project root, affecting the agent's future behavior in that environment.

impeccable