sdlc-ai-assisted-development

SDLC AI Assisted Development

Purpose

Use AI coding agents as supervised engineering workflows, not free-form generators.

Operating Depth

• Lean: produce one concise artifact or decision summary with explicit assumptions.
• Standard: produce the repo artifact, decisions, risks, owners, and validation path.
• Enterprise-ready: add traceability, evidence, auditability, sign-off points, and residual risk.

Workflow

1. Prepare the minimum useful context: repo conventions, specs, architecture, tests, constraints, and risks.
2. Decompose work into bounded tasks with expected artifacts and validation commands.
3. Assign agent roles only when independent perspectives add value.
4. Require spec-to-code traceability, tests, review, and rollback awareness.
5. Use agents/ role definitions for permissions, autonomy level, outputs, gates, and handoffs.
6. Use harnesses/ to coordinate multi-stage autonomous flows and integrations/ to adapt the same flow to Codex, Claude Code, Hermes, OpenClaw, GitHub Actions, MCP, or a custom runtime.
7. Evaluate outputs for correctness, maintainability, security, UX, and scope control.

Expected Inputs

• User request, product context, repository state, or existing SDLC artifacts.
• Relevant files under docs/, tests, configs, architecture notes, and source code when working in a repo.
• Applicable preset from presets/ when the domain changes the risk profile.

Outputs

• Primary artifacts: docs/engineering/ai-development-plan.md, docs/product/requirements-traceability.md, docs/quality/validation-plan.md.
• Clear assumptions, open questions, risks, and owner recommendations.
• Links between product, behavior, architecture, code, tests, release, and production signals where applicable.
• Agent task graph with selected roles, autonomy level, runtime adapter, permissions, gates, and stop conditions when automation is requested.

Guardrails

• Do not allow AI-generated changes that invent APIs, ignore local patterns, skip tests, or refactor outside scope.
• Do not equate high automation with unlimited authority: production, secrets, destructive migrations, payments, dependency additions, and customer data require human approval.
• Prefer existing repo patterns, local conventions, and versioned artifacts over generic advice.
• Keep outputs actionable: decision, rationale, trade-offs, next action, and validation.

Source Guidance

Use the repository sourcing docs when deeper background is needed:

• docs/references/standards-map.md
• docs/references/sourcing-matrix.md
• docs/references/market-skill-benchmark.md
• docs/references/agent-runtime-matrix.md