# Structured Workflow > [!IMPORTANT] > **Work in progress.** The 20 workflow skills are available for review and > experimentation. Project initialization, templates, companion subagents, > hooks, and packaged installation are still being designed. Structured Workflow started as an MCP server in the Sonnet 3.5/3.7 era, when I was trying to answer a question: “How do I get my agents to follow a more structured approach, like a human would, when facing new problems and unknown territory?” This update evolves that idea by bringing in the MYP Design Cycle from my years as a teacher as the core process for building anything with agents. Structured Workflow is a collaborative skillset for agents and humans. It gives coding agents an evidence-producing design cycle as a workflow structure, guiding them through the reasoned process that a product developer, designer, or engineer would use when approaching and solving a problem. It can be lightweight or support detailed analysis and research. It provides a useful default direction while allowing the work to follow new questions, ideas, prototypes, implementation discoveries, and evaluation evidence wherever they lead. The repository currently contains 20 skills: one parent and four criterion skills for each of four phases. It also defines the durable project files that preserve shared understanding and workflow position across sessions. ## The design cycle The default direction is to work through phases A–D, one criterion at a time. Work can move to any criterion or phase that best answers the current question. Research, ideas, exploratory visuals, and evaluation can move naturally between criteria as the current question changes. ### [A — Inquire and Analyze](skills/inquiry-analysis/README.md) - A1 — Explain and justify the need - A2 — Identify and prioritize research - A3 — Analyze prior art - A4 — Develop the Design Brief The phase produces an approved Design Brief supported by inquiry evidence. ### [B — Develop Ideas](skills/developing-ideas/README.md) - B1 — Develop design specifications - B2 — Develop feasible ideas - B3 — Present and justify the chosen design - B4 — Develop planning drawings and diagrams The phase produces an approved Spec that defines the chosen solution. ### [C — Create Solution](skills/creating-solution/README.md) - C1 — Construct a logical plan - C2 — Demonstrate technical skills - C3 — Follow the plan to create the solution - C4 — Justify changes to the design The phase produces approved tickets, a working solution, verification evidence, and justified changes. ### [D — Evaluate](skills/evaluating/README.md) - D1 — Design testing methods - D2 — Evaluate the solution against the design specification - D3 — Explain how the solution could be improved - D4 — Explain the solution's impact The phase produces an evidence-backed verdict, improvements, and an impact explanation. The default direction inside A and B follows criterion order. Criterion C starts with C1, demonstrates C2 while carrying out C3, and uses C4 when creation requires a design change. A complete final evaluation follows D1 through D4. Evaluation is also available throughout the cycle. A Design Brief, Spec, plan, ticket set, prototype, implementation, or claim can be evaluated whenever the work needs an evidence-based judgment. ## How artifacts develop Each phase maintains one authoritative project document: 1. The **inquiry document** preserves A1–A4 evidence and produces the Design Brief. 2. The **developing-ideas document** preserves B1–B4 evidence and completes the Spec. 3. The **creating-solution document** preserves the logical plan, technical skills evidence, creation record, and justified changes. Approved implementation tickets live in the project's issue tracker. 4. The **evaluation document** accumulates reviews and final evaluation evidence. The documents preserve the reasoning behind each handoff. A later phase can load the approved artifact first and retrieve deeper evidence from the owning phase document when needed. ## Durable project context Structured Workflow uses a small set of project files to survive context loss: - **`AGENTS.md`** contains the short, always-loaded workflow rules. - **`workflow-tracker.md`** records the current phase, exact criterion, current question, active artifact, target result, and likely next move. - **`GLOSSARY.md`** preserves the shared meaning of important project terms. - **Phase documents** preserve findings, evidence, decisions, and uncertainty. - **Project templates** define the document locations and reusable expectations, including the project's Definition of Done. The tracker records position rather than findings. Research and reasoning belong in the phase document that owns them. Implementation work belongs in the issue tracker. ## Human-agent collaboration Prefer retrieval-led reasoning over training-led reasoning. The agent begins with the project, its code or materials, direct human knowledge, and primary sources. Material claims are cited where they are recorded. Supported facts, human judgments, inferences, and unknowns remain distinguishable. The agent develops questions from the request, project evidence, industry evidence, contradictions, constraints, and missing understanding. It leads retrieval, review, synthesis, and other bounded work. The human directs intent, priorities, values, and consequential choices. The balance can be even or mostly agent-led, depending on the human and the work. This evidence trail gives the human a concrete basis for inspecting the agent's reasoning, evidence, and uncertainty. ## Repository status The current repository includes: - the always-loaded workflow rules in [`AGENTS.md`](AGENTS.md); - four parent skills and sixteen criterion skills under `skills/`; - four human-facing phase guides; - a project-level [workflow tracker template](skills/workflow-management/workflow-tracker.md). The following surfaces are still planned: - project initialization and document templates; - companion subagent definitions; - continuity and guardrail hooks; - packaged installation and update commands. The skills can be inspected and tested now, but the repository does not yet provide a complete installer. Projects using the workflow must currently place the skills and project files through their agent harness's normal configuration mechanism. ## Repository structure ```text AGENTS.md README.md skills/ inquiry-analysis/ README.md start-inquire-and-analyze/ explain-and-justify-the-need/ identify-and-prioritize-research/ analyze-prior-art/ develop-design-brief/ developing-ideas/ README.md start-develop-ideas/ develop-design-specifications/ develop-feasible-ideas/ present-and-justify-the-chosen-design/ develop-planning-drawings-and-diagrams/ creating-solution/ README.md start-create-solution/ construct-a-logical-plan/ demonstrate-technical-skills/ follow-the-plan-to-create-the-solution/ justify-changes-to-the-design/ evaluating/ README.md start-evaluate/ design-testing-methods/ evaluate-against-the-design-specification/ explain-how-the-solution-could-be-improved/ explain-the-solutions-impact/ workflow-management/ workflow-tracker.md ``` The phase folders organize the skills; they are not additional workflow phases. `workflow-management/` contains cross-phase project resources. ## Sources and acknowledgments Structured Workflow adapts the four criteria of the [IB MYP Design Cycle](https://www.ibo.org/programmes/middle-years-programme/curriculum/design/) for human-agent software work. The criterion structure was checked against the [IB MYP Design Subject Brief](https://www.ibo.org/globalassets/new-structure/brochures-and-infographics/pdfs/myp-brief_design_2015.pdf) and the diagrams published by [Design and Inquiry](https://sites.google.com/view/designandinquiry/myp-design). The workflow and skill design were informed by: - [CodeWithAndreaPro/agentic-coding-toolkit](https://github.com/CodeWithAndreaPro/agentic-coding-toolkit) - [VeryGoodOpenSource/vgv-wingspan](https://github.com/VeryGoodOpenSource/vgv-wingspan) - [VeryGoodOpenSource/vgv-ai-flutter-plugin](https://github.com/VeryGoodOpenSource/vgv-ai-flutter-plugin) - [mattpocock/skills](https://github.com/mattpocock/skills) - [obra/superpowers](https://github.com/obra/superpowers) - OpenAI's Product Design skill pack - [Cline Memory Bank](https://docs.cline.bot/prompting/cline-memory-bank) - [OthmanAdi/planning-with-files](https://github.com/OthmanAdi/planning-with-files) The human-judgment and evidence-trace principles were also informed by Shaw and Nave's research on [cognitive surrender](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646).