vibe-goal
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform extensive shell operations using
git, including managing worktrees, creating/deleting branches, and replaying commits. These operations are core to the skill's function as a developer tool.\n- [REMOTE_CODE_EXECUTION]: The skill requires the execution of repository-specific test suites and type-checking tools ('full suite') to verify code changes before they are integrated.\n- [DATA_EXFILTRATION]: The skill interacts with external 'hosted trackers' (issue management systems) to update ticket statuses and post comments. This interaction is gated by mandatory user previews and explicit approval steps in Stage 4.\n- [PROMPT_INJECTION]: The skill processes untrusted data from external sources, creating a surface for indirect prompt injection.\n - Ingestion points: Data is ingested from user-provided goal descriptions and external issue tracker content fetched during the planning and implementation stages (SKILL.md).\n
- Boundary markers: The skill establishes clear instructional boundaries for sub-agents and uses a ledger to track state, though it does not explicitly define character-level delimiters for all external data.\n
- Capability inventory: The orchestrator has the capability to run shell commands (git), execute repository code (tests), and dispatch sub-agents with specific tasks.\n
- Sanitization: The skill employs a mandatory verification and review process for all sub-agent returns, providing a layer of validation against malicious or incorrect output.
Audit Metadata