repeat-until-settled

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local shell commands, specifically git utilities such as git rev-parse, git status, git diff, and git ls-files, to monitor workspace changes and determine iteration outcomes.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its reliance on untrusted data for control-flow decisions.
  • Ingestion points: Workspace file contents (read during state-capture) and the textual summaries produced by 'inner' skills invoked during the loop.
  • Boundary markers: No explicit delimiters or isolation instructions are used to separate the processing of sub-skill output from the orchestrator's logic.
  • Capability inventory: The orchestrator is capable of invoking any other available skill and performing git/filesystem operations.
  • Sanitization: Summaries from sub-skills are used directly for outcome classification (e.g., detecting 'settled' or 'stalled' states) and are interpolated into new directives during 'Forward escape' cycle resolution.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 02:11 PM
Security Audit — agent-trust-hub — repeat-until-settled