Develop — Multi-Step Workflow

This skill runs in sequential steps. Each step is a self-contained file with its own halt conditions. Load one at a time.

When to invoke

• Phase ≥ 4 (planning complete)
• @tech-lead has approved artifacts/output/05-planning/execution-plan.md
• Worktrees allocated (multi-developer mode, via step 3b) OR single-developer mode
• Implementation is executed by @developer from the approved plan (via worktrees in multi-developer mode)

Prerequisites

• Spec-kernel exists at artifacts/output/03-strategy/ (see spec-kernel-template.md)
• User stories in artifacts/output/03-strategy/user-stories.md
• If a prototype is in scope, the @product-designer + @tech-lead prototype-collaboration.md record exists and prototype_collab_gate.js verify --scope handoff passes for the recorded track and prototype. Do not infer a missing collaboration record from a showcase HTML file.

Mode detection

• Always "create" on first activation.
• Resume is automatic: read stepsCompleted from the output document's YAML frontmatter and jump to the first uncompleted step.

Step loader

1. Read stepsCompleted array from artifacts/output/05-execution/develop-state.md (or start at []).
2. Compute next step = first step NOT in stepsCompleted.
3. Load steps/step-{NN}-*.md.
4. Execute. On completion, append NN to stepsCompleted and re-invoke loader.

Harness adherence (non-negotiable)

• Follow the step sequence exactly. Do NOT skip steps or reorder them.
• Quality gates are not optional. QA runs automatically at step 7 — do NOT ask the user whether to run tests.
• Security and performance gates are auto-decided based on spec content — do NOT ask the user which gates apply.
• Each step file is a contract. Read it fully before executing. Step files override general guidelines.
• If a step halts (test failure, security finding, 2+ review cycles), stop and escalate to @tech-lead. Do NOT proceed past a halt condition.

Step sequence

0. Scope & Decision Anchoring (non-skippable gate) → steps/step-00-scope-and-decision-anchoring.md — step_tracker.js exits 1 if Step 1 begins without a locked track. Intent triage at gate: Clear (C ≥ 0.85) runs the locked track; Ambiguous (0.50–0.85) halts to a 2–3 Track Fork; Trivial (C < 0.50) executes directly. Gate enforced in all step_tracking modes including off (DNA 4).
1. Spec Alignment & Read Check → steps/step-01-spec-alignment.md
2. Architecture (conditional: if ArchitectPhase: true) → steps/step-02-architecture.md 3a. Arch Review → steps/step-03a-arch-review.md 3b. Backlog Prep → steps/step-03b-backlog-prep.md
3. Kanban Activation → steps/step-04-kanban-activation.md
4. Spike (optional) → steps/step-05-spike.md
5. Dev Loop (worktrees from step 3b, multi-developer OR single) → steps/step-06-dev-loop.md 6a. Fidelity Review (coverage contract + 10-class audit before QA — 02o) → steps/step-06a-fidelity-review.md
6. Quality Gates (QA hard gate → security conditional → performance conditional) → steps/step-07-quality-gates.md
7. PM Verification → steps/step-08-pm-verification.md
8. Documentation → steps/step-09-documentation.md
9. Completion → steps/step-10-completion.md

Halt conditions (any one halts the entire flow)

• Spec gap unfilled after 2 CR cycles
• Test failure that doesn't reproduce locally
• Security finding rated Critical or High
• 2 review loops exceeded (escalate to @tech-lead)

State machine integration

At start: node .agents/scripts/orchestrator_state.js status At each step end: node .agents/scripts/orchestrator_state.js step --name {step-name} --status {done/blocked} At end: node .agents/scripts/orchestrator_state.js complete --agent developer --artifact 05-execution/{feature}.md

Memory integration

At start (step-01): Load context before reading any specs:

@memory-controller load developer [develop — {feature name}]

After step-06 (dev loop complete): Write key implementation decisions:

@memory-controller write active-decisions.md
### [CODE] {feature} implementation decisions [date: YYYY-MM-DD] [agent: @developer]
{key decisions: patterns chosen, libraries used, tradeoffs made}
**Status:** active

At completion (step-10): Write session summary — mandatory:

@memory-controller session-write [agent: @developer]
Worked on: {feature name} implementation — steps 01–10 complete
Decisions: {bullet list of key decisions, max 5}
Next step: {launch, iterate, or retro}
Blockers: {any unresolved issues or "none"}

Two surfaces — every step

/develop ends by telling the owner two different things, and they must not be confused:

• Document surface — develop-state.md, test-report.md, coverage-report.md and qa-signoff.md carry the machine record: the verdict, the evidence, the counts. Plain sentence first, then the table.
• Chat surface — the reply carries the plain sentence, the consequence, and the next action. Never a gate token, and never a card field (VERDICT:, Status:, Disposition:). A reader who has never seen this engine must be able to act on the sentence alone (.agents/references/vespyr-dna.md §Two surfaces).

State the consequence, not the label. [CONDITIONAL] is not "conditional" — it is "it ships, but the PDF export fails above 500 rows and @developer owns it, due before launch."

Done when

• develop_completion_check.js check --require-signoff exits 0
• All steps in stepsCompleted
• qa-signoff.md exists with a bracketed GO/CONDITIONAL verdict, and every condition it names carries an owner and a date
• Phase advanced to quality