The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes metadata (names, descriptions, and tags) from external files like stdlib/catalog/skill-catalog.json. This presents a surface for indirect prompt injection where instructions embedded in other skills' metadata could theoretically influence the agent's behavior during the discovery process.
Ingestion points: SKILL.md (bash commands querying local catalog files).
Boundary markers: Absent.
Capability inventory: Read, Grep, Glob, Bash (limited to python and jq binaries).
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill uses bash commands including jq, find, and grep to search the local filesystem (stdlib directory). These commands are restricted to read-only discovery tasks and do not interact with sensitive system files or external networks.
[SAFE]: No evidence of credential exposure, data exfiltration, obfuscation, or persistence mechanisms was found. The skill's operations are consistent with its stated purpose as a meta-tool for library management.

skill-router