The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as find, grep, and gh (GitHub CLI) to search for existing skills. Specifically, it uses user-derived keywords and synonyms directly in commands like grep -iE "keyword|synonym", which could lead to command injection if the generated keywords contain shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest and process SKILL.md files from external sources (GitHub, marketplaces, and web searches).
Ingestion points: Processes SKILL.md files retrieved from local marketplace directories and remote GitHub repositories.
Boundary markers: The skill lacks technical boundary markers (like XML tags) for external content, though it includes a "Vet External Matches" step to manually review instructions.
Capability inventory: The skill possesses shell execution capabilities (find, grep, gh) and web search capabilities.
Sanitization: While Step 4 instructs the agent to manually look for "unexpected shell commands" or "credential handling," there is no automated sanitization or sandboxing for the ingested markdown content.

skill-scout