The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands (cp, mkdir, test, grep) to manage repository structures and bootstrap core templates from its own installation directory. These actions are transparent and aligned with the tool's purpose as a workflow generator.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided specification files (Stage 01) to generate new skill artifacts. While the ingestion of untrusted data constitutes an attack surface for downstream prompt injection, the skill acts solely as a structural builder and does not execute the processed content as instructions during the generation phase.
Ingestion points: Stage 01 reads the spec file path provided as the first argument in SKILL.md.
Boundary markers: The skill uses {{PLACEHOLDER}} syntax in templates but lacks explicit sanitization of spec-file content during interpolation.
Capability inventory: The skill uses Read, Write, Bash, Glob, and Grep tools to generate and validate the new procedure structure.
Sanitization: Content is extracted and mapped to templates; no specific escaping or validation of the input text's safety is performed beyond structural validation.
[DATA_EXPOSURE]: The Discovery stage (Stage 01) scans local skill directories (~/.claude/skills/ and ~/.agents/skills/) to identify candidates for the procedure's dependency list. This access is necessary for the skill's functionality to provide domain-specific tool suggestions.

procedure-builder