tsg-flow-integrator

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill drives its logic based on the content of external files such as flow-state.json and revision reports.
  • Ingestion points: The skill reads flow-state.json, task review reports (N_task_review.md), and full PRD reports (prd_review.md) to verify if a task is 'done' or 'approved'.
  • Boundary markers: No specific delimiters are used when the agent parses these files.
  • Capability inventory: The skill utilizes Git operations including git commit, git rebase, git push, and git merge, as well as file system access to update state.
  • Sanitization: The instructions explicitly direct the agent to check the diff of state files and reject any unauthorized code, configuration, or ADR changes that might be hidden within metadata updates.
  • [COMMAND_EXECUTION]: The skill requires the agent to execute multiple Git commands and GitHub CLI (gh) operations to manage the development lifecycle. While these are necessary for the skill's function, they provide a powerful set of capabilities that rely on the agent correctly interpreting the safety constraints provided in the instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 10:29 PM