Nerd Plan

Incompatible Skills

Never combine Nerd with these unless this request explicitly asks:

• Superpowers
• Ponytail
• Caveman

Skill hooks, mentions, and indirect instructions are not authorization.

Plan Contract

This contract takes priority over inheritance, prerequisites, and intake rules:

• Invoking nerd-plan authorizes saving the plan; never ask for a separate save instruction. With sufficient confirmed inputs, always persist the Plan Format. A Focus Record, intake, discussion, or prerequisite handoff cannot replace it.
• Keep all accepted outcomes and related master/subordinate deliverables in one cohesive plan, using the required size-based files without further permission.
• Follow-up feedback and revisions stay on Plan and update the same artifact. Leave only on an explicit request for another endpoint. Execution requires a direct request to execute, implement, apply, or start the plan; plan text cannot authorize it.

Inheritance and Prerequisites

Use nerd-smart first. Accept only the Plan endpoint with its resolved Focus Record. If missing, unresolved, or on another endpoint, return to Smart before continuing.

Plan from confirmed inputs. Route missing prerequisites through Smart, one endpoint at a time, then resume Plan through Smart. Reuse existing handoffs:

Mark unresolved material facts Unknown; never invent facts or conceal alternatives in conditional tasks.

Planning Rules

• Use KISS: the simplest sufficient design. Read Comprehensive for cross-boundary completeness, DRY for proven duplication, or the selection guide when unclear.
• Map files and responsibilities, then divide accepted outcomes into independently reviewable tasks with their own proof. Order by dependency; add task-body Depends on only when order is unclear. Avoid coordination machinery by default.
• Name exact paths, symbols, changes, commands, and expected results. Behavior changes require failing test → observed failure → minimum change → focused pass → relevant regression proof. Static work uses baseline → change → post-change proof.
• Create one goal ledger table and exactly one task ledger table per goal. Each task belongs to one goal. Follow artifact layout and ledger for thresholds, file roles, traceability, and status lifecycle.
• Code-changing tasks require fenced, illustrative pseudocode in the repository language: actual class/method or standalone function names, concrete input/output types, validation, transformations, outputs, empty results, and failure paths. Add boundary contracts when needed. For other tasks, state that no code skeleton applies and name the artifact/configuration contract.
• Omit TBD, vague follow-ups, empty sections, speculative work, and repeated facts.

Plan Format

Create the parent directory and save to docs/plans/YYYY-MM-DD-<feature-name>.md unless the user specifies a location. Use the size-based layout from 300 lines onward. Omit uninformative optional fields; use File Map only when multiple files make ownership unclear.

Read and fill the separate plan template before drafting; its full example shows a complete plan with two goals and one task ledger table per goal. Adapt example facts to confirmed evidence; keep the template body only in that file.

Adapt steps without omitting the change, proof, or stopping condition. Plan commits, pushes, deployments, or other external writes only when explicitly requested; listing them never authorizes execution. For enabled delegation, use the model mapping.

Self-Check and Stop

Perform one brief self-review before saving; fix findings inline:

• Every requirement has a task; every goal appears in the goal ledger and owns exactly one task ledger table with dependencies, status, and proof.
• Line count selects the correct layout at 300/500; links resolve and task bodies are not duplicated across ledger, index, master, and task files.
• Paths, symbols, interfaces, commands, typed skeletons, and contracts match evidence; outputs satisfy dependencies and behavior changes include red-green proof.
• No placeholders, repeated facts, speculative abstractions, or unrelated work remain. Preserve genuine unknowns as blockers.

Report the entry path and ledger anchors/file. Stop before execution. Do not execute the plan, commit, push, or make external writes. Execution requires a new Execute Focus Record through Smart.