Plan Importer

Plan Importer

Use this skill to turn existing project planning material into a PlanWeave Plan Package.

Command Entry

Resolve the command before writing files:

1. Use a user-provided command if given.
2. Else try global planweave.
3. In the PlanWeave repo, prefer pnpm --filter @planweave-ai/cli planweave.
4. If the repo defines another local script, use that exact entry and show it in the report.

Write examples as <pw> ..., where <pw> is the resolved command.

Use the CLI to resolve the target workspace before writing files:

<pw> init --json
<pw> paths --json
<pw> validate --json

Use the CLI for command discovery, workspace/path resolution, and validation in this workflow. Materialize imports by writing package files directly into the CLI-returned PlanWeave workspace paths.

Workspace

• Run <pw> from the target project root, not from the PlanWeave repo unless PlanWeave itself is the target project.
• Do not create or write ./.planweave inside the target project by hand.
• PlanWeave stores project plans under PLANWEAVE_HOME when set; otherwise defaults to:
  • macOS: ~/.planweave
  • Linux: ~/.planweave
  • Windows: %USERPROFILE%\.planweave
• Read exact workspace paths from <pw> init --json and <pw> paths --json.
• Write only inside CLI-returned PlanWeave workspace paths, such as workspaceRoot, projectGraphPath, and package directories under that workspace.
• For a single-canvas import with an explicit existing target canvas, write only inside that canvas package directory.
• For a single-canvas import without an explicit target canvas, do not overwrite the existing default/current canvas. Create a new formal canvas entry in project-graph.json, allocate a collision-free canvas id, and write the imported package under that canvas package directory.
• For formal multi-canvas plans, write project-graph.json at the returned project/workspace root and write each canvas package only under the package directories named by that project graph.
• Treat CLI-returned project/workspace and package directories as the only writable PlanWeave locations.

Plan File Editing Boundary

• Use CLI/runtime commands for mechanical workspace operations: canvas creation, path allocation, id dedupe, active canvas selection, validation, recovery transactions, and runtime state/results changes.
• Edit Plan Package semantic files directly inside CLI-returned workspace paths: project-graph.json canvas intent and dependencies, each canvas manifest.json tasks/blocks/edges/acceptance/prompt paths, and source prompt Markdown.
• Use narrow CLI edit commands when they exactly express the semantic change; otherwise update the source package files and prompts directly.
• After direct plan edits, run canvas-scoped validation for edited canvases and project validation when project-graph.json, canvas edges, or crossTaskEdges changed.

Import Workflow

1. Scan README, planning docs, ADRs, issues, specs, domain notes, and referenced source files.
2. Record the scanned source list before writing.
3. Extract the PlanWeave package structure: project graph, canvases, tasks, blocks, dependencies, prompt placement, layout hints, and verification strategy.
4. Do not create context nodes. Put goals, requirements, constraints, risks, references, and architecture gates into project/global prompt, task acceptance, task prompt, or block prompt.
5. Run the Plan Quality Gate below before writing. Build a coverage map: each task has concrete acceptance, each block has verifiable done criteria, and key requirements have an explicit prompt placement.
6. Choose an explicit existing canvas, a new single imported canvas, or a formal multi-canvas project graph. If the user did not explicitly ask to replace an existing canvas, create a new canvas instead of writing into default.
7. Run <pw> init --json and <pw> paths --json; use those outputs as the authority for workspace location.
8. Materialize the plan: update project-graph.json when creating a new canvas or multi-canvas plan, then write each canvas manifest.json, task prompts, and block prompts under the declared package directories.
9. Run <pw> validate --json; fix validation errors and weak importer-created coverage.
10. Output a Plan Import Report listing source docs, command entry, project graph path when present, package paths, prompt placement, canvas strategy, review strategy, and validation result.

Plan Quality Gate

• Map every authoritative goal to a task, block, acceptance item, or prompt; flag omitted, weakened, or incorrectly deferred requirements.
• Do not accept a demo subset as complete delivery unless the user explicitly scoped it that way.
• Identify core objects and trace create, structure, validation, transform, state, storage, consumer, side effects, final output, failure, retry, rollback, and manual intervention.
• Keep schema, types, APIs, CLI flags, events, files, and prompt inputs/outputs consistent across producers and consumers.
• Contract-changing tasks must cover callers, tests, fixtures, docs, and migrations when relevant; do not hide missing contracts with fallback, default values, any, or mock-only paths.
• Model the real execution order: parallel tasks must be independent, sequential gates must be explicit, and each canvas must map to a stage, capability area, subsystem, workflow, or parallel work group.
• Default imported multi-canvas plans to crossTaskEdges: [].
• For multi-canvas imports, canvas-level order must be encoded as formal project graph edges. Before adding a cross-task edge, first move tightly coupled tasks into the same canvas or promote stage-wide ordering to a canvas dependency edge.
• Use explicit crossTaskEdges only for sparse, irreducible task blockers between otherwise cohesive and independently executable canvases, and record why task relocation or a canvas edge would be incorrect.
• Treat dense or repeated cross-task edges between the same canvases, stage-wide ordering expressed as task edges, and task edges redundant with canvas order as a canvas-boundary defect; repartition the tasks or promote the dependency before writing.
• Do not rely on project/global prompt prose as the only source of canvas order or cross-canvas blockers.
• Reject fake completion such as schema with no runtime use, API with no caller, UI with no behavior, config never read, provider abstraction without live client, queue without consumer, file path without file, dry-run without live path, or fixture-only testing.
• Cover errors, retry, cancellation, timeout, permission failure, partial success, external outage, failed human review, and recovery when the domain needs reliable execution.
• Every block must name concrete validation: commands, tests, output artifacts, observable state changes, or end-to-end flows.
• Complex blocks must encode architecture boundaries, test location, config/env handling, README or .env.example updates when applicable, and real provider vs mock/dry-run expectations in done criteria.
• Do not copy other projects' skills, bootstrap rules, or prompt conventions into this plan unless the target repository explicitly requires them.
• Separate plan defects from PlanWeave toolchain defects in the report.

Prompt Placement

• PlanWeave Global Prompt / Project Prompt: cross-cutting rules, architecture constraints, reference files, coding standards, shared risks.
• Task Prompt: task-local context, acceptance rationale, dependencies, files likely touched.
• Block Prompt: exact execution instructions, validation commands, output/report expectations.
• Do not write rendered prompt output back into source prompt files. Rendered prompts come from <pw> prompt <ref> and are derived artifacts.
• Do not leave block prompts empty. If a block needs no separate detail, say that it inherits task/project context and list the concrete done condition.

Task And Block Granularity

• Do not split for the sake of splitting. Split by data flow, ownership boundary, dependency, risk, or independently verifiable acceptance.
• Merge tasks that are too small to claim, test, or report independently.
• Use implementation blocks for executable work and encode explicit verification in done criteria, validation commands, or review gates.
• Add review blocks only for complex code, cross-contract changes, database/schema migration, provider integration, security/privacy, architecture changes, or high-risk user-visible behavior.
• Do not add review blocks for simple docs, config tweaks, local copy edits, or low-risk single-file changes unless the user asks.

Multi-Canvas Strategy

• For small plans, use one package canvas.
• For large plans, especially 100+ tasks/nodes, split by stage, capability area, subsystem, workflow, or owner into task canvases.
• Keep each canvas scannable, usually 10-30 tasks when the source plan allows it.
• Materialize large or multi-stage imports as a formal project-graph.json plus one manifest.json per canvas.
• Use project graph canvas dependency edges for canvas-level order.
• Keep crossTaskEdges empty unless a sparse, irreducible blocker passes the Plan Quality Gate.
• Keep single-canvas manifest.json semantics unchanged; do not place cross-canvas edges inside a canvas manifest.
• The project/global prompt may explain strategy, but it is not the authority for dependency enforcement.
• Different canvases are not automatically parallel; derive mandatory order from project graph edges and crossTaskEdges. Record possible overlap with parallel.sharedResources; it is advisory and must not be presented as preventing execution.

Project Graph Shape

When writing a formal multi-canvas plan, include:

• project-graph.json at the PlanWeave project/workspace root.
• canvases entries with stable ids, titles, package directories, state/results paths when required by the schema, and no invisible legacy-only canvas records.
• canvas dependency edges for stage, capability, subsystem, or workflow order.
• explicit crossTaskEdges only for documented sparse exceptions that cannot be represented by task placement or canvas dependency edges.
• no context nodes, feedback nodes, runtime state, or layout-only graph mirrors.

After writing, validate with <pw> validate --json; project graph schema/read/compile diagnostics such as missing canvas refs, missing cross-task refs, and cycles must be fixed before reporting success.

Block Shape

Each block needs id, type, title, prompt, depends_on, optional parallel.sharedResources hints, done criteria, validation, and report expectations. If review is justified, add R-001 after implementation blocks with review.required: true and a clear review prompt.

Rules

• Treat project-graph.json (when present), each canvas manifest.json, source prompts, and task/block prompt files as plan content source of truth.
• If the user provides existing package-shaped files, verify their structure and materialize them directly into the target workspace files.
• Do not create feedback block types; feedback is runtime state.
• Do not write implementation state into the manifest.
• Use state.json only for runtime task/block/feedback state.
• Do not create runtime graph mirrors, .plan/, SQLite stores, HTTP APIs, Docker services, or MCP servers.
• Do not use the Plan Import Report as a substitute for manifest edges, task acceptance, or prompt content.