skills/identity16/skills/make-loop/Gen Agent Trust Hub

make-loop

Pass

Audited by Gen Agent Trust Hub on Oct 30, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill's primary function is to create and manage local configuration files (loop.md, rules.json) within the user's data directory (~/.local/share/identity16-skills/). This is standard behavior for managing state.
  • [SAFE]: The skill includes explicit safety guardrails, including a 'Never' section in the loop definition that specifically forbids dangerous actions like moving money, posting publicly, or entering passwords.
  • [SAFE]: The instructions mandate a human-in-the-loop requirement, stating that 'a loop with no review step is not a loop' and requiring user approval for the definition and any automated rules.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow that processes data from external sources to generate proposals. This creates a theoretical attack surface where malicious content in the source data could attempt to influence the agent's behavior. However, the mandatory human review step and the 'zero-trace contract' (which prevents the skill from writing to target project files) serve as significant mitigations.
  • Ingestion points: External data sources defined in the Source section of the loop definition.
  • Boundary markers: Not explicitly defined in this skill; the agent is expected to follow the instructions provided in the loop definition.
  • Capability inventory: File-write operations to the user's local data directory for configuration and state tracking.
  • Sanitization: Not specified in the instructions; the system relies on user review of proposals.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 30, 2026, 03:38 PM