skills/dpolivaev/spec-loop/plan-task/Gen Agent Trust Hub

plan-task

Pass

Audited by Gen Agent Trust Hub on May 11, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill enforces a rigorous governance model called a 'Constitution' which prevents the agent from making autonomous executable changes without explicit user approval. This methodology significantly reduces the risk of accidental or malicious code modifications.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and apply project-specific instructions from files like AGENTS.md. While this constitutes a potential attack surface for indirect prompt injection, the risk is mitigated by the 'PLAN' phase restrictions which limit the agent to editing non-executable artifacts until a human provides approval.
  • Ingestion points: AGENTS.md, project glossary files (glossary.adoc, glossary.md), and existing task files.
  • Boundary markers: The skill relies on the 'Constitution' logic rather than explicit string delimiters to separate instruction layers.
  • Capability inventory: File system read/write (limited to task files and documentation during PLAN) and shell command execution for research purposes.
  • Sanitization: No explicit sanitization of ingested content is mentioned, but the workflow requires the agent to self-check and the user to approve all designs before implementation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 11, 2026, 05:54 PM
Security Audit — agent-trust-hub — plan-task