The Agent Skills Directory

[SAFE]: The skill's general operations are safe. It performs local tool discovery using standard Node.js modules and does not engage in network communication or unauthorized file access.
[SAFE]: Analysis of the indirect prompt injection surface (Category 8):
Ingestion points: The query parameter in scripts/main.cjs receives untrusted input from the agent during tool discovery searches.
Boundary markers: Absent. The skill handles input raw for tokenization, though the resulting matches are returned as structured JSON which limits injection potential into the agent context.
Capability inventory: The skill is restricted to reading the local .claude/config/tool-manifest.json file using the fs module to retrieve tool metadata.
Sanitization: Input is processed via a whitelist-based tokenization regex /[^a-z0-9_:-]+/ in scripts/main.cjs, which effectively filters out special characters and potential malicious instructions.

tool-search