The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands via Bash to delete files identified as stale or redundant. The execution is limited to specific internal directories (.claude/) and is consistent with the skill's stated maintenance purpose.
[PROMPT_INJECTION]: The skill processes memory files which could contain untrusted data, representing a surface for indirect prompt injection.
Ingestion points: Memory files located in .claude/agent-memory/, .claude/projects/, and .claude/memory/.
Boundary markers: None; the skill parses raw markdown content and frontmatter from the identified files.
Capability inventory: File deletion via Bash (rm), directory searching via Glob, and index updates via Write.
Sanitization: None; logic relies on deterministic regex matching and word-overlap ratios to identify files for removal.

dream