The Agent Skills Directory

[SAFE]: The skill functions as a local diagnostic tool. It scans component directories (agents, skills, rules) and configuration files (.mcp.json, CLAUDE.md) to perform token counting and identify redundant content. These operations are performed locally and align with the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it ingests and processes user-defined files (agent definitions and rules) to generate its report.
Ingestion points: Reads contents of agents/*.md, skills/*/SKILL.md, rules/**/*.md, and CLAUDE.md.
Boundary markers: None explicitly defined for the ingested content.
Capability inventory: Limited to local file read access and terminal output for reporting; no network, file-write, or shell execution capabilities are present.
Sanitization: Not specified; however, since the skill only performs quantitative analysis (counting words and lines), the risk of the agent executing malicious instructions embedded in the analyzed files is minimal.

context-budget