merge-orchestrator

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill includes a diagnostic "Reporting workflow" that instructs the agent to capture and share a debug payload to public GitHub issues. This payload contains sensitive environment information, specifically absolute file system paths (collected via repoRoot and worktreeList), which exposes the internal directory structure and local username of the host machine to an external third party.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input to drive local file system mutations, creating an indirect prompt injection surface.
  • Ingestion points: sourceBranch, targetBranch, and taskId parameters defined in the merge_orchestrate action schema within SKILL.md.
  • Boundary markers: Absent. The instructions do not specify the use of delimiters or ignore-instructions warnings when processing these branch names.
  • Capability inventory: The skill utilizes the exarchos MCP server to execute powerful local operations including git merge, git reset, git checkout, and git worktree, which can modify repository state or reveal metadata.
  • Sanitization: Absent. No validation or escaping is prescribed for branch names, which could be exploited for command injection if the underlying tool handler lacks sufficient sanitization.
  • [COMMAND_EXECUTION]: The core functionality of the skill relies on orchestrating local Git command execution to manage merges, recovery points, and worktree preflights. These commands involve significant file system access and mutation capabilities.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 05:19 AM