This repository is an intro/tutorial project for running the front end of a
design thinking sprint with agentic software practices. It turns
docs/brief.md, supporting documents, and the ordered resources in
exercises/ into evidence-tracked framing artifacts for later ideation and
prototyping.
- Treat
docs/brief.mdas the primary challenge input. - Treat Markdown files under
exercises/as the canonical ordered exercise definitions. The numeric prefix controls execution order, but each exercise must be analyzed as an independent action before synthesis. - Treat HTML and PDF files under
exercises/as optional local source material. They are ignored by Git and must not be required by the runtime. - Treat
human_input/anddocs/user_input/as human-in-the-loop evidence drop zones. If evidence is absent, generate the appropriate collection template instead of inventing field data. - Treat
outputs/graph/as append-only provenance storage. Never rewrite or delete claims, evidence, or conflict history. - Keep ideation and prototyping out of scope. The system may prepare inputs for those phases but must not propose final solutions as if they were validated.
- Use least privilege for every agent. An agent receives only the exercise input, prior approved outputs, and graph query access needed for its task.
- Preserve evidence provenance. Every takeaway must point to source document, exercise, agent, confidence, validation status, and graph claim IDs.
- Record structured logs for stage, agent, input manifest, output manifest, graph mutations, warnings, and validation decisions.
- Do not log hidden chain-of-thought. Log concise public rationale, assumptions, evidence references, and decision summaries.
- Prefer explicit schemas and small modules over broad, multi-mode functions.
- Handle inconsistency as design material unless stronger evidence invalidates a claim. Deprecated claims remain in the graph with invalidation edges.
The final compiler should produce:
- framed challenge with named audience and correct altitude;
- defined audiences and role distinctions;
- stakeholder persona mappings with pains, gains, and jobs to be done;
- landscape and constraint facts;
- opportunity reframings, usually in How Might We form;
- project plan for later ideation and prototyping phases.
Use config/project.yaml as the repo-specific configuration source. Use the
agent profiles in agents/*.md, the MCP manifest in mcp/manifest.yaml, and
the repo-local skill in skills/design-sprint-agentic/.