plan

• SKILL.mdcore · 31,765 bytes
• SKILL.mdcore · 31,765 bytes
• SKILL.mdcore · 31,765 bytes
• agents/openai.yamlasset · 375 bytes
• references/breadboarding.mddoc · 1,810 bytes

# Knowledge plan

## Contract

| Field | Bound contract |
|---|---|
| Trigger | User commits to a direction and asks to plan, brief, structure, research, or operationalize it; or asks to score a plan against a quality bar, break a goal into dependency-ordered tasks, shape a pitch, render a plan visually, enumerate and diagram a decision field, audit or tune a supplied plan, have a plan-mode enforcement hook intercept a plan review, or lock a greenfield architecture and interface contracts. |
| Authority | Reversible local: default modes write only named local plan artifacts (the plans/ file, the score-mode plan file and review report, and the visual-mode page in the diagrams directory); review audit writes its verdict page and review tune writes its question registry and hook configuration; waterfall writes named architecture and interface contract files. Research is read-only. Storm, breakdown, and shape write no files. Rollback is deleting or reverting those artifacts or restoring prior configuration. No remote, VCS-history, credential, paid, published, or deployed mutation. |
| Side effect | Default writes plans/{type}-{descriptive-name}.md, adding a date on collision; score writes a plan file and review report; visual writes one self-contained HTML page and opens it; review writes a verdict page, atomically persists tuning configuration, or renders the current tuning state for inspect; waterfall writes a locked architecture document for distribution; storm returns an exhaustive field and diagram; breakdown and shape return chat output only. |
| Done | Default: type and tier classified, research checks run, the user acknowledges the context brief, and the file leads with the type-correct answer plus sourced metrics, questions, and references. Score: 5/5 on all six dimensions or a named blocker. Breakdown: every task carries a checkable acceptance criterion and dependency order, and the user approves. Shape: a five-ingredient pitch at fat-marker altitude or a single verdict. Visual: nine sections in order ending in an observable acceptance checklist. Storm: every plausible option is enumerated, cross-referenced, and diagrammed before a choice. Review audit: every plan item has an accuracy classification and final approve, revise, or reject verdict; review tune: the requested state is atomically persisted and re-read, or the inspect request renders the dual-track profile and preference map. Waterfall: the architecture and interface contracts are complete, human-confirmed, locked, and distributed. |

## Inputs

- Default and plan-authoring modes require a user commitment to a direction and a descriptive name for the plan. Mode-specific inputs below replace that requirement where stated.