Cook

Canonical end-to-end implementation command. Routing (complexity, risk lane, who-decides) comes from workflows/routing-card.md — read that one file, not the full specs. Development Rules (workflows/development-rules.md) holds the Verification Gate + Voice; orchestration-protocol.md is the full orchestration spec if a case falls outside the card. Do not re-derive any of them here.

<input>$ARGUMENTS</input>

Input auto-detection

Inspect <input>:

• References an existing plan (a path under ./plans, or words like "the plan / latest plan") → Execute mode: skip planning, go straight to Implement using that plan.
• A task/feature description → Full mode: Research → Plan → Implement.
• Empty → ask the user what to build (one question).

Modifiers (flag-literalism)

A flag is active ONLY if its literal token appears in <input>:

• --auto → autonomous "trust me bro": no per-step pauses; auto-handle review cycles; still honor the Verification Gate. Default (absent) = interactive: confirm at plan approval + before commit.
• --no-test → skip the Testing step (use sparingly).
• --no-worktree → skip session worktree isolation (workflows/worktree-session.md §1-§2); cook runs in the current checkout. Default (absent) = worktree isolation ON.
• Parallel is auto-detected, not a flag: if the plan has a file-ownership matrix, launch cook-worker agents in parallel per the matrix (tier via model param + brief; cap, overlap-detection, wave scheduling, barrier + fail-at-barrier: see workflows/parallel-execution.md); otherwise implement sequentially. Log "detected: parallel/sequential".
• --no-hoh → opt OUT of the default continual-improvement flow. Without this flag, step 4 delegates to /vk-hoh-run (HoH is default ON) instead of direct phase-by-phase cook-worker execution. HoH runs its own plan→dev→qa loop on the workspace (Runtime.check build+execute, evidence-bound acceptance, issue-ledger, per-stage commits) for T loops (default --loops 3). Cook keeps the outer lifecycle (worktree → research → plan → [HoH inner loop] → test → review → finalize). --no-test is NOT forwarded to HoH — HoH QA always verifies; --no-test only skips cook's post-HoH tester step. HoH reads its loop budget from --loops; falls back T=3. With --no-hoh: original phase-by-phase cook-worker execution (legacy flow).

Lifecycle

0. Enter Worktree — session-level isolation per workflows/worktree-session.md §1 (nested-guard → derive name via /vk-worktree slug rules → EnterWorktree → copy plans/<dir> in Execute mode). Skipped entirely if --no-worktree.
1. Clarify (Full mode) — ask probing questions, one at a time, until the request is clear. Run Risk-Lane Intake (workflows/routing-card.md §2): mark applicable flags, determine lane (tiny/normal/high-risk). Then scope what each subagent receives per the phase×lane matrix + token budgets in workflows/context-rules.md — pass by file path, never paste full files. Activate needed skills from the catalog.
2. Research (Full mode, if complex) — parallel researcher agents (reports ≤150 lines, cited). Use /vk-scout to locate files.
3. Plan (Full mode) — planner agent creates a progressive-disclosure plan (plans/YYYYMMDD-HHmm-name/plan.md + phase-XX-*.md). Interactive: get user approval before implementing. --auto: proceed.
4. Implement — Default (HoH ON, unless --no-hoh): delegate step 4 to /vk-hoh-run <plan-path (or S=spec)> --loops N [--interactive if not --auto]. HoH runs the inner plan→dev→qa loop and yields reports/final.md (loops run, gaps, regressions, issue-ledger). Cook resumes at step 5 with HoH's final evidence as input. With --no-hoh: main agent (or parallel cook-worker if matrix present) implements phase by phase. ui-ux-designer for frontend per ./docs/design-guidelines.md. Run typecheck/compile after each phase.
5. Test (unless --no-test) — tester runs the suite under the Evidence Gate (paste real output). On failure: debugger root-cause → fix → retest. Loop bound: 3 cycles → escalate. No fake data. A build-deploy-e2e pre-test sub-sequence may run first per the activation rule in workflows/build-deploy-e2e-loop.md §Composability.
6. Review — code-reviewer under the Findings Gate (quote file:line, confidence 1-10). Fix criticals, retest. --auto: auto-approve at score ≥ 9.5 & 0 criticals; else escalate after 3 cycles.
7. Finalize — on approval, project-manager + docs-manager (parallel) update plan status / ./docs / ./docs/project-roadmap.md. Then commit (interactive: confirm via /vk-git-cm; --auto: auto-commit when tests pass) → auto-PR per workflows/worktree-session.md §3 (interactive: confirm once before opening the PR; --auto: auto-push + auto-PR under the same trigger). Default cleanup: ExitWorktree({action:"keep"}) after PR (skip if step 0 skipped worktree creation).
8. Onboarding & Report — guide setup if needed (one question at a time); concise final report; list unresolved questions at the end. Include worktree branch/path and PR link (if created) in the final report.

Notes

• Assets: prefer SVG/CSS/icons; analyze images/screenshots with the Read tool natively; web screenshots via chrome-devtools. No built-in raster generator — ask the user for raster images.