The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from a WORKFLOW_QUESTIONS.md report, which serves as the basis for automated source code modifications and command execution. While it has strong verification rules, it represents an attack surface for indirect prompt injection.\n
Ingestion points: The WORKFLOW_QUESTIONS.md file is loaded and parsed in Step 1.\n
Boundary markers: The skill explicitly instructs the agent to treat the report as "evidence, not truth" and "hypotheses, not confirmed facts" (Rule 6).\n
Capability inventory: The skill is authorized to write to docs/workflows/新项目开发工作流/ (Step 9) and execute the bump-workflow-version.py script (Step 10B) and trellis CLI commands (Step 12).\n
Sanitization: The skill mandates a "Temp-project-first verification" (Rule 3) and a "Truth Precheck" (Step 2B) where all findings must be verified against actual filesystem contents before any action is taken.\n- [COMMAND_EXECUTION]: The skill executes the bump-workflow-version.py script and interacts with the trellis CLI (e.g., trellis continue, trellis finish-work) to manage project tasks. These are local operations within the expected environment of the workflow being managed.

workflow-repair