GitHub

Feature lifecycle

When the job is a product feature, Waykit routes work through specialist roles: grilling, PRD/bet (when the value is unproven), stories, spec, TDD, cross-functional quality, audit, telemetry and release - then confirm or kill flagged bets. Language and framework profiles load on top of that once the stack is known. The orchestrator stays in the parent chat. Isolation, sequential and readonly-audit roles on the subagent allowlist launch as host subagents by default; they write COMPLETE/BLOCKED to the handover (subagent launch). Run wk agents status before routing. Set WK_SUBAGENTS=0 in the host’s shell, then confirm mode: skills-only, to keep those roles as skills in the parent. EDD (alpha) is how you prove tool routing in CI; it lives inside TDD when the change is a prompt or a tool schema. Product bets: hypothesis-driven development. Bugs: hypothesis-driven debug.

StepRuntime
Grilling, PRD, storiesSkills in the parent
Spec, TDD (gear 1+2), XFNHost subagents (or skills if wk agents status says skills-only)
Security / hex-drift auditReadonly host subagents
Telemetry, release, pruneSkills in the parent

The sequence and Task prompt live on subagent launch, not here.

  1. Grilling: If the idea is still mushy, interview until the decision frontier is clear — including contract vs bet.
  2. PRD / bet: For bets, write problem, belief, leading indicator, timebox, kill criteria, cheapest experiment, and flag plan (agent-prd). Tiny contracts may skip this.
  3. Stories then spec: INVEST tickets (Hypothesis block on bets; flags in Notes), then Gherkin including flag off/on/kill when flagged.
  4. TDD: Inventory the behavior catalog (both flag states), then gear 1 (domain) and gear 2 (thin adapters) in the same agent-tdd subagent. Claim the Linear issue In Progress and assign the host agent first (linear ticket workflow). Stay on main, uncommitted; output the conventional commit subject with the ticket id. EDD lives here when the change is a prompt or tool schema. Cursor/Claude TDD Guard hooks (wk tdd-guard install) block production writes until a failing test run is recorded.
  5. XFN: Green the apply rows (accessibility, load, security) or skip with a reason — including flag-on surfaces in scope.
  6. Audit: Security and architecture-drift checks, then pre-commit.
  7. Telemetry and release: Map SLOs (agent-telemetry) and the bet’s leading indicator in PostHog (agent-posthog). Ship with a conventional PR title, record flag expiry and rollback.
  8. Close the loop: After the timebox, measure that leading indicator in PostHog (wk mcp posthog --install). Then confirm (default on, prune flag) or kill (flag off, prune slice) via agent-user-stories and agent-prune. Do not add a separate product-insights role. Turning live PostHog signals into backlog work is the two-session loop on product signal intake — link only; do not copy the procedure here. Git crime-scene ranking → Linear is complexity hotspots §8 — same gate, no extra skill. SonarQube issues are sonarqube findings — triage, skip kit disagreements, file one typed ticket after restore default, no extra skill. Coding-agent tokens, cost, and OTel traces are coding-agent observability — community SigNoz Docker on the operator machine, SigNoz Cloud ingest for Cloud Agents, wk mcp signoz or hosted MCP, wk eval for right results, no extra skill. CI, RUM, Lighthouse, and scout alerts that should become typed Linear tickets are quality loops — hygiene, allowlist, then draft PR, no extra skill.

Orchestrator skill · Coding philosophy · Quality loops · EDD guide (alpha) · Hosts · Subagent allowlist

Markdown source