Epic Workflow Skill

This skill guides the agent through an end-to-end process for planning large features (Epics) and generating their respective GitHub issues. It follows a strict Requirement Drill ➔ Documentation ➔ Decomposition pipeline.

When to use

• When starting a new feature (Epic) that requires research and multi-issue planning.
• Specifically triggers when the user says "start a new epic", "plan this feature", or mentions "epic workflow".

Context Management (Checkpointing)

Long-running workflows like Epics can exceed the model's context window. To maintain efficiency and accuracy:

1. Phase Checkpointing: After EVERY phase, summarize the findings into the skill's own tmp directory: epic-workflow/tmp/epics/<epic-name>/context/.
   • After /grill-me: Save to grill-summary.md.
   • After research: Save to research-notes.md.
2. "Stateless" Resumption: If the session gets too long, instruct the user that you are "checkpointing". Next time you start, simply view_file on these summaries instead of re-processing the entire chat history.
3. Artifact-as-Truth: Treat the generated files (PRD, Plan, Tasks) as the persistent source of truth. Read the latest file version from disk rather than relying on previous chat messages.
4. Subagent Delegation: For Epics that span multiple domains or require deep research, PROACTIVELY spawn subagents via browser_subagent or generic subagent tools. Delegate specialized work like UI research, API deep-dives, or schema verification to them to keep your main context focused on the high-level orchestration.