agent-patterns

Pass

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The documentation describes patterns that create an Indirect Prompt Injection surface by design.
  • Ingestion points: Patterns such as 'Fungible Swarm' and 'Batch Workflows' (detailed in references/delegation-patterns.md) involve sub-agents reading and acting upon project files, task boards, and arbitrary data items.
  • Boundary markers: The skill suggests 'Context Bundling' and 'Manifest of Objective' (in references/orchestration.md) but does not provide specific instructions for using delimiters or warnings to ignore commands embedded in processed data.
  • Capability inventory: Templates describe agents with high-privilege tools including Bash, Write, Edit, and WebFetch.
  • Sanitization: The skill does not mention sanitizing or validating ingested data content before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill provides documentation on high-privilege patterns including the use of the Bash tool and lifecycle hooks that execute local shell scripts (e.g., './scripts/run-linter.sh' in references/sub-agent-configuration.md). It also details the 'bypassPermissions' mode which allows agents to skip standard permission checks.
  • [EXTERNAL_DOWNLOADS]: Research-oriented agent patterns described in the documentation involve the use of 'WebFetch' and 'WebSearch' tools to access and retrieve data from external sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 27, 2026, 04:32 AM