grant-decision-authority

Pass

Audited by Gen Agent Trust Hub on Oct 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill directs the agent to execute the grep utility to scan the local repository for specific tracking markers. This is a read-only operation used for internal bookkeeping of decisions and does not involve network access or sensitive file modification.
  • Evidence: grep -rn "TEMPORARY AGENT" . is used to enumerate open loans of authority.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting information from project documentation and issue trackers, which could potentially contain malicious instructions. The risk is mitigated by a mandatory human-in-the-loop review process.
  • Ingestion points: The agent reads AGENTS.md, CLAUDE.md, specifications, Architectural Decision Records (ADRs), and project tickets to determine existing repository decisions.
  • Boundary markers: The agent is required to wrap every temporary decision in a verbatim marker string (TEMPORARY AGENT...) and document it in a structured PR table.
  • Capability inventory: The agent can search the local tree using grep, write markers to files, and interact with related project management skills for PR creation.
  • Sanitization: Every decision made under this authority must be explicitly approved or rejected by a human in a "settle-up" session before the markers are removed and the code is allowed to merge.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 1, 2026, 01:25 AM