Software Development Workflow
Use this as an orchestrator for substantial end-to-end software work. Do not load or apply every specialized skill automatically; select only the lanes the task actually needs.
Route the Task
- Unvalidated product idea or disputed direction: use
product-strategy-review; use product-requirements after the direction is accepted but behavior or scope still needs definition.
- Consequential design choice: use
architecture-decision and, for sensitive systems, threat-modeling.
- Existing-code changes: apply
implementation-standards and change-safety.
- Unfamiliar or consequential change surface: use
codebase-mapping before planning implementation.
- Dependency or persisted-data changes: use
dependency-management or database-migrations.
- Tests and bug fixes: use
testing-standards, debugging-checklist, and test-and-fix-loop as appropriate.
- Broad verification: use
repo-quality-audit; use subagents only for independent, bounded lanes.
- Review: use
code-review and security-baseline in proportion to risk.
- Developer-facing API, CLI, SDK, library, platform, or documentation journey: use
developer-experience-review in plan or live mode.
- Visual direction still open: use
design-exploration; after approval, use frontend-design, frontend-quality, accessibility-review, visual-regression, or webapp-testing only for the implementation and verification lanes needed.
- MCP server work: use
mcp-builder.
- Substantial specs, RFCs, ADRs, or runbooks: use
technical-doc-coauthoring; apply documentation-style to the final form.
- User-facing or operational changes: update documentation and observability where relevant; use
changelog-generator for release communication.
- Issue-backed delivery: use
issue-to-pr-workflow; otherwise prepare with git-workflow, pr-preparation, and release-checklist.
- Newly deployed web release: use
canary-monitoring for read-only baseline comparison after the authorized deployment.
- Cross-session continuity or durable repository knowledge: use
context-handoff for work-in-progress state and project-learnings for verified long-lived facts.
- Periodic delivery/process improvement: use
engineering-retrospective; use postmortem instead for a specific stabilized incident.
Execution Gates
- Establish scope, success criteria, constraints, repository instructions, and current state.
- Make the smallest coherent implementation and keep unrelated user work untouched.
- Add or update tests that prove the requested behavior.
- Run focused checks, then broader verification proportional to risk.
- Review the final diff, update affected documentation, and report residual risk.
Do not infer permission to stage, commit, push, open issues or pull requests, deploy, publish, migrate production data, or communicate externally. Require explicit authorization for each external mutation.
1---2name: software-development-workflow3description: Coordinate substantial software work through the necessary requirements, design, implementation, testing, review, documentation, and release stages.4---56# Software Development Workflow78Use this as an orchestrator for substantial end-to-end software work. Do not load or apply every specialized skill automatically; select only the lanes the task actually needs.910## Route the Task1112- Unvalidated product idea or disputed direction: use `product-strategy-review`; use `product-requirements` after the direction is accepted but behavior or scope still needs definition.13- Consequential design choice: use `architecture-decision` and, for sensitive systems, `threat-modeling`.14- Existing-code changes: apply `implementation-standards` and `change-safety`.15- Unfamiliar or consequential change surface: use `codebase-mapping` before planning implementation.16- Dependency or persisted-data changes: use `dependency-management` or `database-migrations`.17- Tests and bug fixes: use `testing-standards`, `debugging-checklist`, and `test-and-fix-loop` as appropriate.18- Broad verification: use `repo-quality-audit`; use subagents only for independent, bounded lanes.19- Review: use `code-review` and `security-baseline` in proportion to risk.20- Developer-facing API, CLI, SDK, library, platform, or documentation journey: use `developer-experience-review` in plan or live mode.21- Visual direction still open: use `design-exploration`; after approval, use `frontend-design`, `frontend-quality`, `accessibility-review`, `visual-regression`, or `webapp-testing` only for the implementation and verification lanes needed.22- MCP server work: use `mcp-builder`.23- Substantial specs, RFCs, ADRs, or runbooks: use `technical-doc-coauthoring`; apply `documentation-style` to the final form.24- User-facing or operational changes: update documentation and observability where relevant; use `changelog-generator` for release communication.25- Issue-backed delivery: use `issue-to-pr-workflow`; otherwise prepare with `git-workflow`, `pr-preparation`, and `release-checklist`.26- Newly deployed web release: use `canary-monitoring` for read-only baseline comparison after the authorized deployment.27- Cross-session continuity or durable repository knowledge: use `context-handoff` for work-in-progress state and `project-learnings` for verified long-lived facts.28- Periodic delivery/process improvement: use `engineering-retrospective`; use `postmortem` instead for a specific stabilized incident.2930## Execution Gates31321. Establish scope, success criteria, constraints, repository instructions, and current state.332. Make the smallest coherent implementation and keep unrelated user work untouched.343. Add or update tests that prove the requested behavior.354. Run focused checks, then broader verification proportional to risk.365. Review the final diff, update affected documentation, and report residual risk.3738Do not infer permission to stage, commit, push, open issues or pull requests, deploy, publish, migrate production data, or communicate externally. Require explicit authorization for each external mutation.