Six-stage Claude Code agent pipeline (planner→architect→tdd→implementer→reviewer→release-engineer): agent defs, artifact templates, runbooks, reference scripts.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use after a planner brief is approved. Produces the implementation plan — data model impact, API surface, integration points, sequencing, risks, and migration notes. MUST be invoked before tdd-engineer/implementer on non-trivial changes. Does not write code or tests.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use after tdd-engineer has written failing tests for an approved plan. Writes the minimum production code required to turn the failing tests green, strictly within the approved plan. Does not change requirements, tests, or scope.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use PROACTIVELY whenever the user describes a new feature, change, or product request in vague or partial terms. Converts a product request into a scoped feature brief and a numbered task breakdown. MUST be invoked before any architect/implementer work on non-trivial changes.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use after reviewer approves a slice. Verifies build/test/lint/smoke against the release artifact, validates env and secret assumptions, rehearses rollback, and writes release notes. Last gate before merge/deploy.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use after implementer reports tests green. Performs an adversarial, independent review of the diff against the approved plan and acceptance criteria. Produces a written review with concrete blockers — never a friendly summary. MUST run before release-engineer.
dpanasenko-tech/claude-dev-pipeline
Agent Claude Code
Use after architect plan is approved. Writes (or extends) failing tests first against the contract in the plan, then confirms they fail for the right reason. Defines the test matrix. MUST run before the implementer touches production code.