The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate specification documents in .claude/specs/ which are intended for use by downstream agents.
Ingestion points: User's initial ambiguous requests and subsequent dialogue responses are ingested into the agent context in SKILL.md.
Boundary markers: The skill uses a structured Socratic dialogue but lacks explicit delimiters or instructions for the agent to ignore potentially malicious embedded commands when interpolating user goals into the final Markdown specification.
Capability inventory: The skill utilizes the Bash tool for directory management (mkdir -p) and file verification (test -f), and the Write tool to save the resulting document. The generated specs are explicitly intended to be passed to other agents like agent-composer.
Sanitization: While the skill applies sanitization to the filename (slug generation), it does not explicitly sanitize the user-provided content before writing it to the document, creating a trust boundary issue where a user could embed instructions for the next agent in the pipeline.

deep-clarify