The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes user-supplied problem statements to generate directory slugs for persistence.
Ingestion points: The $ARGUMENTS variable in SKILL.md captures the user's decision or problem statement.
Boundary markers: There are no explicit delimiters or instructions to treat the user input as untrusted data.
Capability inventory: The skill performs file system reads (Read, Grep, Glob) and writes history files to .agent-history/reasoning/<slug>/.
Sanitization: The slug generation logic does not sanitize for path traversal characters such as '..', which could allow the agent to write files to unintended locations on the file system.

sc-hypothesize