The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to search for and read SKILL.md files from remote GitHub repositories and web search results. This process creates a surface for indirect prompt injection, where malicious instructions embedded in a public skill file could potentially influence the agent's logic during the vetting and ranking phases.\n
Ingestion points: Remote skill files retrieved via gh search and web searches as described in Step 3 and Step 4.\n
Boundary markers: No technical delimiters or isolation techniques are specified for handling the ingested external content, although the agent is instructed to manually vet the content for safety.\n
Capability inventory: The skill utilizes shell commands (find, grep) and the GitHub CLI (gh) for search operations.\n
Sanitization: There is no programmatic sanitization or filtering of the content retrieved from external sources; the skill relies entirely on the agent's manual review.\n- [COMMAND_EXECUTION]: The skill uses local shell commands like find and grep to scan the ~/.claude/skills directory and the GitHub CLI (gh) for remote repository metadata searches. These operations are restricted to search functionality and are consistent with the skill's intended purpose.

skill-scout