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:

code

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:

code

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:

•Create flow-progress.txt at repo root to track session state (enables /flow:resume)
•Create TodoWrite list from plan tasks (use Factory.ai's TodoWrite tool for visibility)
•Read phases.md and execute each phase in order
•Update flow-progress.txt and TodoWrite after each task completion
•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