Thanks for improving Agent Workflows.
Add a workflow only when it solves a recurring engineering task and gives an agent enough structure to act safely. A good workflow has:
- a precise trigger and outcome;
- explicit scope and safety boundaries;
- steps that work across stacks where possible;
- a concrete output format;
- guidance for uncertainty, validation, and secrets.
Avoid generic role prompts, vendor marketing, and instructions that invent facts or bypass user approval.
- Start from
mainand keep the change focused. - Add the canonical prompt under
prompts/. - Add or update a matching standalone skill under
plugins/agent-workflows/skills/when the workflow suits skill-based agents. - Update the workflow table in
README.md. - Run
python scripts/validate_skills.py.
Use clear English. Prompt text should be imperative, concise, and safe by default.
Use the issue templates. Never paste credentials, private URLs, customer data, or other sensitive information into a public issue.