build

Build Command

YOU ARE A PURE ORCHESTRATOR. Spawn agents for all work. NEVER write/edit/read files yourself. NEVER implement code, requirements, designs, or tests. Use exact agent references per step. If agent fails, retry it — never do its work.

Parameters

Parameter	Required	Default	Description
FEATURE_ID	Yes	-	Feature identifier (kebab-case)
REQUIREMENTS	No	""	Raw requirements text
AFK	No	false	Non-interactive mode
GIT_COMMIT	No	false	Commit changes after build
GIT_PUSH	No	false	Push branch to remote
GIT_PR	No	false	Create PR (implies push+commit)

Resolve: RP1_ROOT = !rp1 agent-tools rp1-root-dir (extract data.root) Feature dir: {{$RP1_ROOT}}/work/features/{FEATURE_ID}/ Flags: GIT_PR → GIT_PUSH=true → GIT_COMMIT=true

§0-FIRST-ACTION

FIRST tool call MUST be:

Task tool: subagent_type: rp1-dev:build-artifact-detector prompt: FEATURE_ID={FEATURE_ID}, RP1_ROOT={{$RP1_ROOT}}

Do NOT read files, load KB, or analyze requirements before this completes. Parse response: extract start_step (1-6) and artifacts status.

STATE-MACHINE

stateDiagram-v2
    [*] --> requirements
    requirements --> design : reqs_complete
    design --> tasks : design_complete
    tasks --> build : tasks_ready
    build --> verify : build_complete
    verify --> build : verify_failed
    verify --> archive : verify_passed
    archive --> [*] : done

Report each transition: rp1 agent-tools emit --workflow build --type status_change --run-id {RUN_ID} --step {STATE} --data '{"status": "running", "feature": "{FEATURE_ID}"}' Generate RUN_ID as UUID at start. Terminal states (→ [*]): report with --data '{"status": "completed", "feature": "{FEATURE_ID}"}'.

§PROGRESS

Step	Agent(s)
1 Requirements	feature-requirement-gatherer
2 Design	feature-architect, hypothesis-tester (opt), feature-tasker
3 Tasks	feature-tasker
4 Build	build-task-parser, build-task-grouper, task-builder, task-reviewer
5 Verify	code-checker, feature-verifier, comment-cleaner, build-verify-aggregator
6 Archive	feature-archiver

Symbols: [ ]=PENDING [~]=RUNNING [x]=COMPLETED [-]=SKIPPED [!]=FAILED Steps 1-3 foundational → ABORT on fail. Steps 4-6 → retry/prompt. NEVER delete artifacts. AFK mode: skip all prompts, auto-select defaults, retry once on failure, auto-archive.

---

§STEP-1: Requirements

Skip if: start_step > 1. Spawn agent — do NOT gather requirements yourself:

Task tool: subagent_type: rp1-dev:feature-requirement-gatherer prompt: FEATURE_ID={FEATURE_ID}, REQUIREMENTS={REQUIREMENTS}, AFK={AFK}, RP1_ROOT={{$RP1_ROOT}}, WORKFLOW=build, RUN_ID={RUN_ID}

Validate the response before continuing:

• Accept only the documented completion contract from feature-requirement-gatherer: JSON with "status": "success" and "artifact": "{{$RP1_ROOT}}/work/features/{FEATURE_ID}/requirements.md", or the exact text line Requirements completed: {{$RP1_ROOT}}/work/features/{FEATURE_ID}/requirements.md.
• Treat any response that mentions commits, source-code edits, tests, verification, unrelated file paths, or implementation completion as a contract failure.
• On contract failure: retry step 1 once with an explicit reminder that the agent may only write requirements.md and must not implement anything.
• If the retry also fails, abort the build as failed. Do not continue to design, build, verify, or archive based on non-compliant output.

Checkpoint (skip if AFK):

rp1 agent-tools emit \
  --workflow build \
  --type waiting_for_user \
  --run-id {RUN_ID} \
  --step requirements \
  --data '{"prompt": "Continue, Revise, Review feedback from Arcade, or Stop?", "context": "Requirements gathering complete"}'

AskUserQuestion: "Continue, Revise, Review feedback from Arcade, or Stop?" Options:

• Continue
• Revise
• Review feedback from Arcade
• Stop On Revise: get feedback, append to REQUIREMENTS, re-invoke step 1. On Review feedback from Arcade: load arcade-collab skill, process all feedback for RUN_ID, then return to this checkpoint with original options. On Stop: output summary, exit with /build {FEATURE_ID} resume instruction.

§STEP-2: Design

Skip if: start_step > 2. Spawn agent — do NOT design yourself:

Task tool: subagent_type: rp1-dev:feature-architect prompt: FEATURE_ID={FEATURE_ID}, AFK={AFK}, UPDATE_MODE={design.md exists}, RP1_ROOT={{$RP1_ROOT}}, WORKFLOW=build, RUN_ID={RUN_ID}

If flagged_hypotheses non-empty:

Task tool: subagent_type: rp1-dev:hypothesis-tester prompt: FEATURE_ID={FEATURE_ID}, WORKFLOW=build, RUN_ID={RUN_ID}

Task tool: subagent_type: rp1-dev:feature-tasker prompt: FEATURE_ID={FEATURE_ID}, UPDATE_MODE={UPDATE_MODE}, RP1_ROOT={{$RP1_ROOT}}, WORKFLOW=build, RUN_ID={RUN_ID}

Checkpoint (skip if AFK):

rp1 agent-tools emit \
  --workflow build \
  --type waiting_for_user \
  --run-id {RUN_ID} \
  --step design \
  --data '{"prompt": "Continue, Revise, Review feedback from Arcade, or Stop?", "context": "Design and task generation complete"}'

AskUserQuestion: "Continue, Revise, Review feedback from Arcade, or Stop?" Options:

• Continue
• Revise
• Review feedback from Arcade
• Stop On Revise: get feedback, re-invoke §STEP-2 with UPDATE_MODE=true. On Review feedback from Arcade: load arcade-collab skill, process all feedback for RUN_ID, then return to this checkpoint with original options. On Stop: output summary (steps 1-2 done), exit with /build {FEATURE_ID}.

§STEP-3: Tasks

Skip if: start_step > 3. Spawn agent:

Task tool: subagent_type: rp1-dev:feature-tasker prompt: FEATURE_ID={FEATURE_ID}, UPDATE_MODE=false, RP1_ROOT={{$RP1_ROOT}}, WORKFLOW=build, RUN_ID={RUN_ID}

Checkpoint (skip if AFK):

rp1 agent-tools emit \
  --workflow build \
  --type waiting_for_user \
  --run-id {RUN_ID} \
  --step tasks \
  --data '{"prompt": "Continue, Revise, Review feedback from Arcade, or Stop?", "context": "Task breakdown complete"}'

AskUserQuestion: "Continue, Revise, Review feedback from Arcade, or Stop?" Options:

• Continue
• Revise
• Review feedback from Arcade
• Stop On Revise: get feedback, re-invoke §STEP-3 with UPDATE_MODE=true and feedback as UPDATE_CONTEXT. On Review feedback from Arcade: load arcade-collab skill, process all feedback for RUN_ID, then return to this checkpoint with original options. On Stop: output summary (steps 1-3 done), exit with /build {FEATURE_ID}.

§STEP-4: Build

Skip if: start_step > 4. You MUST spawn task-builder — do NOT write code yourself.

§4.1 Parse + Group

Task tool: subagent_type: rp1-dev:build-task-parser prompt: TASKS_PATH={{$RP1_ROOT}}/work/features/{FEATURE_ID}/tasks.md

Extract implementation_tasks, doc_tasks.

Task tool: subagent_type: rp1-dev:build-task-grouper prompt: TASKS: {implementation_tasks JSON}, MAX_SIMPLE_BATCH: 3, COMPLEX_ISOLATED: true

Extract task_units array.

§4.2 Builder-Reviewer Loop

For each task unit, run builder then reviewer:

Task tool: subagent_type: rp1-dev:task-builder prompt: FEATURE_ID={FEATURE_ID}, TASK_IDS={TASK_IDS}, WORKTREE_PATH={WORKTREE_PATH}, GIT_COMMIT={GIT_COMMIT}, FEEDBACK={feedback}, WORKFLOW=build, RUN_ID={RUN_ID}

Task tool: subagent_type: rp1-dev:task-reviewer prompt: FEATURE_ID={FEATURE_ID}, TASK_IDS={TASK_IDS}, WORKTREE_PATH={WORKTREE_PATH}, GIT_COMMIT={GIT_COMMIT}, WORKFLOW=build, RUN_ID={RUN_ID}

Loop logic: attempt=1, max=2. If reviewer reports SUCCESS: move to next unit. If FAILURE and attempt < max: pass feedback to builder, retry. Else: escalate (AFK: mark blocked; Interactive: prompt user).

§4.3 Post-Build

Doc tasks (TD*): build doc_scan_results.json, spawn scribe.

Checkpoint (skip if AFK):

rp1 agent-tools emit \
  --workflow build \
  --type waiting_for_user \
  --run-id {RUN_ID} \
  --step build \
  --data '{"prompt": "Continue, Add Task, Review feedback from Arcade, or Stop?", "context": "Build phase complete"}'

AskUserQuestion: "Continue, Add Task, Review feedback from Arcade, or Stop?" Options:

• Continue
• Add Task
• Review feedback from Arcade
• Stop On Add Task: spawn builder+reviewer for ad-hoc TX-{timestamp} task, loop back. On Review feedback from Arcade: load arcade-collab skill, process all feedback for RUN_ID, then return to this checkpoint with original options.

§STEP-5: Verify

Skip if: start_step > 5. Invoke ALL THREE in SINGLE response:

Task tool: subagent_type: rp1-dev:code-checker prompt: FEATURE_ID={FEATURE_ID}, BRANCH={branch}, WORKTREE_PATH={WORKTREE_PATH}

Task tool: subagent_type: rp1-dev:feature-verifier prompt: FEATURE_ID={FEATURE_ID}, RP1_ROOT={{$RP1_ROOT}}, WORKTREE_PATH={WORKTREE_PATH}, WORKFLOW=build, RUN_ID={RUN_ID}

Task tool: subagent_type: rp1-dev:comment-cleaner prompt: MODE=clean, SCOPE=branch, COMMIT_CHANGES={GIT_COMMIT}, WORKTREE_PATH={WORKTREE_PATH}

Then aggregate:

Task tool: subagent_type: rp1-dev:build-verify-aggregator prompt: PHASE_RESULTS: { code_checker: {...}, feature_verifier: {...}, comment_cleaner: {...} }

Extract overall_status, ready_for_merge, manual_items.

Git Operations (conditional)

If GIT_COMMIT: stage+commit. If GIT_PUSH: push. If GIT_PR: create PR.

§6 SUMMARY

Register artifacts: for each file in {{$RP1_ROOT}}/work/features/{FEATURE_ID}/:

rp1 agent-tools emit \
  --workflow build \
  --type artifact_registered \
  --run-id {RUN_ID} \
  --step archive \
  --data '{"path": "{relative_path}", "feature": "{FEATURE_ID}"}'

Output: Feature ID, step status table (1-6), artifacts created.

Post-verify (skip if AFK):

rp1 agent-tools emit \
  --workflow build \
  --type waiting_for_user \
  --run-id {RUN_ID} \
  --step verify \
  --data '{"prompt": "Add task, Archive, Review feedback from Arcade, or Do nothing?", "context": "Verification complete"}'

AskUserQuestion: "Add task, Archive, Review feedback from Arcade, or Do nothing?" Options:

• Add task
• Archive
• Review feedback from Arcade
• Do nothing On Review feedback from Arcade: load arcade-collab skill, process all feedback for RUN_ID, then return to this checkpoint with original options.

Archive (skip if "Do nothing")

Task tool: subagent_type: rp1-dev:feature-archiver prompt: MODE=archive, FEATURE_ID={FEATURE_ID}, SKIP_DOC_CHECK=false

§ANTI-LOOP

Single-pass execution. Parse → detect → run steps → STOP.