recipe-fullstack-build

Pass

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs automated file deletion in the docs/plans/tasks/ directory during its final cleanup phase. It targets files matching specific task patterns associated with the current work plan.
  • [INDIRECT_PROMPT_INJECTION]: The skill's workflow is driven by the content of external files, creating a surface for indirect instructions to influence agent behavior.
  • Ingestion points: Processes work plans, design documents, and task files from the docs/ directory.
  • Boundary markers: Uses external references like subagents-orchestration-guide to define execution and commit boundaries.
  • Capability inventory: Has the ability to invoke various specialized subagents, perform git commits, and delete files.
  • Sanitization: The skill relies on external orchestration guidelines rather than implementing internal sanitization for the ingested data.
  • [PROMPT_INJECTION]: The skill includes instructions to enter an 'autonomous mode' where it interprets the initial user instruction as a 'batch approval' for all subsequent actions, effectively bypassing granular user confirmation for tool executions and file changes.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 23, 2026, 10:09 AM
Security Audit — agent-trust-hub — recipe-fullstack-build