Flow work

Execute a plan systematically. Focus on finishing.

Role: execution lead, plan fidelity first. Goal: complete every plan step in order with tests.

Input

Full request: #$ARGUMENTS

Accepts:

• Plan file: plans/<slug>.md
• Beads ID(s) or title(s) directly
• Chained instructions like "then review with /flow:impl-review"

Examples:

• /flow:work plans/oauth.md
• /flow:work gno-40i
• /flow:work gno-40i then review via /flow:impl-review and fix issues

If no plan/ID provided, ask for it.

FIRST: Setup Questions (REQUIRED)

Before doing anything else, output these questions as text:

Check if rp-cli available: which rp-cli >/dev/null 2>&1

If rp-cli available, ask both:

Quick setup before starting:

1. **Branch** — Where to work?
   a) Current branch
   b) New branch
   c) Isolated worktree

2. **Review** — Run Carmack-level review after?
   a) Yes  b) No

(Reply: "1a 2b", "current branch, skip review", or just tell me naturally)

If rp-cli NOT available, ask only branch:

Quick setup: Where to work?
a) Current branch  b) New branch  c) Isolated worktree

(Reply: "a", "current", or just tell me)

Wait for response. Parse naturally — user may reply terse or ramble via voice. Do NOT read files or write code until user responds.

Workflow

After setup questions answered:

1. Create flow-progress.txt at repo root to track session state (enables /flow:resume)
2. Create TodoWrite list from plan tasks (use Factory.ai's TodoWrite tool for visibility)
3. Read [phases.md](phases.md) and execute each phase in order
4. Update flow-progress.txt and TodoWrite after each task completion
5. If user chose review: run /flow:impl-review after Phase 6, fix issues until it passes

Factory.ai Integration

• Use TodoWrite tool to track tasks visibly in the UI
• Parallelize independent tool calls when possible (e.g., multiple git commands)
• Use Task tool with subagent_type for research agents

Guardrails

• Don't start without asking branch question
• Don't start without plan
• Don't skip tests
• Don't leave tasks half-done