sdlc-flow

name	sdlc-flow
description	Use when a request spans multiple lifecycle phases or the correct next phase is unclear.
targets	["*"]

SDLC Flow

The plugin is an optional graph of phases. Each phase is a skill that routes to one agent role and emits an artifact for its immediate receiver. This skill explains where to start, which dependencies apply, and how the phases connect.

Read references/core-ontology.md when reasoning about system entities, relationships, cardinalities, lifecycle edges, authority, traceability, evidence, or provider projections. The ontology controls relationships; project glossaries control term definitions.

Host routing contract

When the host exposes the named phase agents and you are the parent/orchestrator, delegate the selected phase to its named owner and pass the complete prior artifact. When already executing as that owner, or when named-agent delegation is unavailable, run the phase locally. Never delegate the same phase recursively. The artifact contract is portable; automatic delegation is a host capability, not something Rulesync can synthesize.

Lifecycle artifact contract

Use lifecycle-documentation for every phase output and input gate. Select the smallest applicable IEEE/ISO information item, write it in STE-derived controlled English, preserve stable IDs, and end with the required handoff block. A receiving phase must reject an incomplete artifact instead of guessing what the producing phase meant.

The default path: idea → shipped

Select only the phases the work needs. Run selected phases in dependency order; each handoff passes one accepted artifact to one immediate receiver.

1. feasibility-analysis (→ researcher) — scout an unfamiliar problem space. Start here when the approach is unknown. Produces options, constraints, and trade-offs.
2. requirements-gathering (→ product-owner) — turn a vague or high-blast-radius request into a stress-tested specification. Start here when what to build is unclear.
3. planning (→ project-manager) — break the spec into a plan of milestones with dependencies, risks, and an explicit out-of-scope set. Produces a plan.
4. design (→ architect) — settle interfaces, contracts, exact files, and trade-offs. Produces a technical design. For user-facing work, pair with visual-design (→ graphic-designer) for layout, composition, states, and accessibility.
5. implementation (→ engineer) — build it test-first in vertical slices. Produces tested, bounded changes. Commit those changes when the task includes commit authority.
6. testing (→ tester) — adversarially falsify the change by executing it across edge cases, scale, and security. Produces a verdict with reproducible failing cases.
7. quality (→ quality-engineer) — assure the change beyond execution: verification & validation, review/inspection, and the quality attributes. Produces a quality verdict.

Operate & evolve (the loop)

Shipping isn't the end — the lifecycle is a loop. What you learn in production re-enters at debugging-and-error-recovery or requirements-gathering.

• deployment (→ release-engineer) — ship safely with rollout, smoke check, and rollback.
• operations (→ sre) — run the live service: observability/SLOs, incident response, runbooks, and continuity. Routes learnings back into the loop.
• documentation (→ technical-writer) — document what actually shipped.
• maintenance (→ maintainer) — fix, simplify, and reduce debt without changing behavior.

Cross-cutting concerns

Some concerns run through the phases, not as a single step — build them in, don't inspect them in at the end:

• Quality — the engineer writes the code's tests test-first; the tester adds adversarial tests; the quality-engineer assures V&V, reviews, and quality attributes.
• Security — threats in requirements, least-privilege in design, security-and-hardening in construction, attacked in testing. Not a final checklist.
• Traceability — give each requirement a stable ID and carry it into design, tasks, and tests so the link stays greppable.
• Visual verification — for any UI change, the graphic-designer inspects the rendered result. Screenshots provide evidence for visual claims; executable tests provide evidence for behavior.
• Actionable handoffs — use actionable-communication so every phase leads with its artifact or blocker, gives compact evidence, and names the next phase only when work remains.
• Lifecycle documents — use lifecycle-documentation to select the correct SRS, plan, design, configuration, test, quality, V&V, user, review, audit, release, or operations artifact and apply the same controlled-English and traceability rules at every phase boundary.
• Simplifying insight — use find-simplifying-insight when rules and exceptions proliferate and the underlying model may be wrong. Require a novel prediction and falsification test; do not force a paradigm shift when the current model survives.

On-ramps

Situations that enter the lifecycle at the smallest applicable phase:

• A bug or unexpected behavior → debugging-and-error-recovery (root-cause it), then implementation to fix it test-first (write a regression test first).
• A performance problem → performance-optimization (measure → fix → guard).
• A security concern → security-and-hardening.
• A UI / visual change → visual-design (design and verify the rendered result).
• Authoring a new frontend surface (landing page, app UI, design system, restyle) → frontend-design (committed art direction, signature element, token-driven system), then visual-design to verify.
• Messy but working code → maintenance.

Standalone / user-invoked

Reach for these methods directly, outside the phase graph:

• interrogate — relentless interrogation to stress-test any plan, design, spec, or approach (the base interrogation the requirements phase builds on).
• interrogate-with-docs — same interrogation, but challenges the plan against the project's documented domain model and updates CONTEXT.md / ADRs inline.
• deslop — strip AI-generated slop from a recent change.