goal-driven-batch

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's prerequisite check may prompt the user to install dependencies using npx skills add FuDesign2008/open-skills -g. This command downloads and installs additional skills from the author's own official repository.
  • [COMMAND_EXECUTION]: The orchestrator executes shell commands for various operations, including using git for branch isolation (via git-worktree-discipline) and the openspec CLI for architectural validation and traceability during the batch run.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes task cards and repository files which are external inputs. It mitigates potential injection risks through a 'frozen decision' model where requirements are deeply interviewed and confirmed by a human before execution, alongside budget caps and automated validation checks that stop the run if the environment deviates from the approved state.
  • Ingestion points: Reads task cards from the .goal-driven/ directory and existing repository files during the consumption-entry check.
  • Boundary markers: Uses structured markdown templates for task cards and a frozen decision ledger to guide the child engine.
  • Capability inventory: Can execute Bash scripts, manage git branches, and write to the local filesystem.
  • Sanitization: Implements a 'consumption-entry check' to re-validate frozen constraints against the current repository state before every task run, ensuring that changes in the environment do not bypass the original human approval.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 12:08 PM
Security Audit — agent-trust-hub — goal-driven-batch