Work

Execute tasks from a plan file with checkpoint tracking and verification.

Usage

/phx:work .claude/plans/user-auth/plan.md
/phx:work .claude/plans/user-auth/plan.md --from P2-T3
/phx:work --skip-blockers
/phx:work  # Resumes most recent plan

Arguments

• <plan-file> -- Path to plan file (optional, auto-detects recent)
• --from <task-id> -- Resume from specific task (e.g., P2-T3)
• --skip-blockers -- Continue past blocked tasks
• --continue -- Resume IN_PROGRESS plan from checkboxes

Iron Laws (NON-NEGOTIABLE)

1. NEVER auto-proceed to /phx:review or any next workflow phase -- always ask the user what to do next
2. AUTO-CONTINUE between plan phases -- when Phase N completes, immediately start Phase N+1. Do NOT stop or ask for permission between phases. Only stop at BLOCKERS or when ALL phases are done.
3. Plan checkboxes ARE the state -- [x] = done, [ ] = pending. No separate JSON state files. Resume by reading the plan.
4. Verify after EVERY task -- never skip verification
5. Max 3 retries then BLOCKER -- don't keep retrying forever
6. Stage specific files -- never use git add -A or git add .
7. Read scratchpad BEFORE implementing -- scratchpad has dead-ends and decisions that prevent rework. Step 2 is not optional.
8. Clarify ambiguous tasks -- ask the user rather than guessing when a plan task's intent is unclear

Step 1: Research Decision

For plans with >3 tasks, ask the user:

This plan has {count} remaining tasks across {count} phases.

1. Start working -- Begin immediately (familiar patterns)
2. Quick research -- Read source files first (~10 min)
3. Extensive research -- Web search + docs (~30 min)

Skip for plans with 3 or fewer simple tasks -- just start.

Split warning: Plans with >10 tasks risk 2-3 context compactions. Suggest splitting via /phx:plan if not already.

Step 2: Check Context (MANDATORY)

Read scratchpad and compound docs before writing any code. Skipping this causes rework — scratchpad captures dead-ends and decisions from planning that prevent taking wrong paths.

# Read full scratchpad — it's short and has critical context
cat .claude/plans/{slug}/scratchpad.md 2>/dev/null
# Check compound docs for solved patterns
grep -rl "KEYWORD" .claude/solutions/ 2>/dev/null

Apply findings: skip dead-ends, follow decisions, reuse patterns. If a task's intent is ambiguous, ask the user before implementing rather than guessing — corrections are expensive.

Step 3: Load, Create Task List, and Resume

Read plan file, count [x] (completed) vs [ ] (remaining). Find first unchecked task by [Pn-Tm] ID.

Create Claude Code tasks from ALL unchecked plan items using TaskCreate. This gives real-time progress visibility in the UI:

For each unchecked `- [ ] [Pn-Tm] Description`:
  TaskCreate({
    subject: "[Pn-Tm] Description",
    description: "Full task details from plan",
    activeForm: "Implementing: Description"
  })

Already-checked items ([x]): skip, don't create tasks for them. Set up blockedBy dependencies between phases (Phase 2 tasks blocked by Phase 1 tasks).

With --from P2-T3: Skip to that specific task.

See references/resume-strategies.md for all resume modes.

Step 4: Execute Tasks

For each unchecked task (- [ ] [Pn-Tm][agent] Description):

1. Start task: TaskUpdate({taskId, status: "in_progress"})
2. Route by [agent] annotation (see references/execution-guide.md)
3. Implement the task
4. Verify: mix format + mix compile --warnings-as-errors (at phase end, also run mix test <affected> — see tiers below)
5. Complete task: Mark checkbox [x] on pass, append implementation note inline, AND TaskUpdate({taskId, status: "completed"}). Example: - [x] [P1-T3] Add user schema — citext for email, composite index on [user_id, status] This survives context compaction; the plan is re-read on resume.
6. On failure: retry up to 3 times, then create BLOCKER and write DEAD-END to scratchpad (see error-recovery.md)

Parallel groups: Tasks under ### Parallel: header spawn as background subagents. See references/execution-guide.md for spawning pattern, prompt template, and checkpoint flow.

Verification tiers:

• Per-task: mix format + mix compile --warnings-as-errors (when Tidewave available, also check get_logs :error after edits)
• Per-phase: above + mix test <affected> + mix credo --strict
• Per-feature (Tidewave): behavioral smoke test via project_eval (create record, fetch, verify -- see execution-guide.md)
• Final gate: mix test (full suite)

Linter note: The PostToolUse hook checks formatting but does NOT modify files. Run mix format explicitly during verification steps or before committing.

Step 5: Completion

Summarize results with AskUserQuestion:

Implementation complete! {done}/{total} tasks finished. {count} files modified across {count} phases.

Options: 1. Run review (/phx:review) (Recommended), 2. Get a briefing (/phx:brief — understand what was built), 3. Commit changes (/commit), 4. Continue manually.

With blockers: list them, offer Replan (/phx:plan), Review first (/phx:review), or Handle myself.

If blockers remain, auto-write HANDOFF to scratchpad:

### [HH:MM] HANDOFF: {plan name}
Status: {done}/{total} tasks. Blockers: {list}.
Next: {first unchecked task ID and description}.
Key decisions: {brief list from this session}.

This gives a fresh session context beyond just checkboxes.

NEVER auto-start /phx:review or any other phase.

Step 6: Check for Additional Plans

After completion, check for other pending plans:

ls .claude/plans/*/plan.md 2>/dev/null

If pending plans exist, inform the user. Do NOT auto-start.

Integration

/phx:plan → /phx:work (YOU ARE HERE) → /phx:review → /phx:compound
                 ↑ ASK USER before each transition

References

• references/execution-guide.md -- Task routing, parallel execution, verification
• references/resume-strategies.md -- Resume modes and state persistence
• references/file-formats.md -- Plan and progress file formats
• references/error-recovery.md -- Error handling and blockers
• references/harness-patterns.md -- Critic-refiner pattern for debugging loops