The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard command-line tools such as git, grep, rg (ripgrep), and find to perform repository analysis and metadata extraction. These operations are scoped to the local environment and intended for project maintenance.
[DATA_EXPOSURE]: The skill is designed to read source code, entity definitions, and existing documentation to synthesize feature specifications. This data access is essential for its primary function and does not involve external exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes untrusted content from the codebase to generate documentation.
Ingestion points: The skill reads repository files, including source code, templates, and documentation (e.g., SKILL.md, references/author.md).
Boundary markers: The skill relies on strict template structures and a "tech-free" prose mandate, but does not explicitly implement instruction-delimiting markers for the data it processes.
Capability inventory: The agent can write files to the project directory and execute shell commands for code searching.
Sanitization: No specific content sanitization or instruction-filtering mechanisms were identified.
[REMOTE_CODE_EXECUTION]: The skill references a local Python script (.claude/scripts/code_graph) for code analysis. No patterns of downloading or executing scripts from remote, untrusted sources were found.

spec