signals-scout-skills-store
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill audits third-party agent skills which are retrieved via the
skill-getandskill-file-gettools. This creates an indirect prompt injection surface where a processed skill could contain hidden instructions designed to influence the behavior of the scout agent. - Ingestion points: Untrusted content is ingested from the skill store's bodies, descriptions, metadata, and bundled files (SKILL.md and referenced files).
- Boundary markers: The skill includes a clear defensive section ('Untrusted content') explicitly instructing the agent to treat all ingested fields as data for analysis and never as instructions to follow.
- Capability inventory: The agent possesses reporting capabilities (
emit_report,edit_report) and access to a persistent scratchpad for storing memory and audit states. - Sanitization: Instructions require the agent to redact credentials and quote only sanitized, short snippets of ingested content in its reports.
- [EXTERNAL_DOWNLOADS]: The skill periodically fetches updated auditing rules from official Anthropic repositories, Claude platform documentation, and established industry specifications.
- Evidence: Fetches data from
https://agentskills.io/specification,https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices, andhttps://raw.githubusercontent.com/anthropics/skills/main/skills/skill-creator/SKILL.md.
Audit Metadata