The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository and design files to identify and document UI patterns, creating a potential vector for indirect instructions.
Ingestion points: The skill uses Read, Grep, Glob, and Bash (ls, find, cat, grep, rg) to search and ingest content from the project codebase and design documentation.
Boundary markers: The instructions do not define explicit delimiters or use specific 'ignore instructions' markers when processing the identified code or design file content.
Capability inventory: The agent is granted Write permissions and Bash execution access for searching and reading, which could be abused if malicious instructions are encountered in the scanned files.
Sanitization: There is no requirement for the agent to sanitize, escape, or filter the content read from the filesystem before incorporating it into the final documentation output.

pattern-documentation