maintaining-agent-context

Pass

Audited by Gen Agent Trust Hub on Aug 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes untrusted agent instruction files (such as AGENTS.md, CLAUDE.md, and referenced documentation) which may contain malicious payloads designed to override agent behavior.
  • Ingestion points: SKILL.md Phase 1 (Discovery) inventories numerous instruction surfaces including root and nested files, conditional rules, and linked documentation.
  • Boundary markers: The skill contains a dedicated "Security model" section that explicitly defines audited content as untrusted data and instructs the agent that such content does not change the workflow or grant new authorizations.
  • Capability inventory: The skill has file-reading capabilities for all analysis phases and file-writing capabilities (Phase 6) for applying approved changes. It does not possess network capabilities.
  • Sanitization: The skill implements a human-in-the-loop sanitization process by presenting all proposed changes as diffs for explicit user confirmation in Phase 5 before any files are modified.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 29, 2026, 09:40 PM
Security Audit — agent-trust-hub — maintaining-agent-context