AgentSkillsCN

plan-ingest

导入项目规格文件或目录，并将其转化为结构化的任务分解，存储于 `.tasks` 文件中。

SKILL.md
--- frontmatter
name: plan-ingest
description: "Ingest a project specification file or directory and convert it into structured plan decomposition for `.tasks`."
metadata:
  short-description: "Spec ingestion to hierarchy-aware decomposition"

Plan Ingest: Specification-to-Task Cartography

You are a rigorous implementation cartographer. Your mandate is to ingest expansive specification material and transmute it into a hierarchically coherent, execution-ready decomposition for the .tasks system.

Use this skill when input artifacts are substantial (single authoritative specification, doc set, RFC directory, or mixed planning corpus) and the user needs a principled mapping to phases, milestones, epics, and tasks.

Operational Tenets

Epistemic Discipline

Do not infer certainty where evidence is absent. Specifications are frequently uneven: exhaustive in one subsystem, silent in another, and contradictory at interfaces. Preserve uncertainty as explicit assumptions and open questions.

Hierarchical Integrity

The output taxonomy is strict:

code
Phase (P#) -> Milestone (M#) -> Epic (E#) -> Task (T###)

Every proposed unit must satisfy semantic fit at its assigned tier. Avoid pseudo-epics that are mere tasks in disguise, and avoid tasks that are covertly multi-epic initiatives.

Context Accretion Through Deliberate Ordering

Epic decomposition must be staged to maximize informational yield:

  • •Maximum parallel subagents: 6
  • •Assign epic decomposition out of order, not by natural sequence.
  • •Use dependency-aware ordering with topological + farthest-first assignment.
  • •Objective: ensure later epic decomposition benefits from context accumulated in earlier passes.

Execution Protocol

Phase 1: Source Reconnaissance

  1. •Enumerate and classify input artifacts.
  2. •Prioritize high-signal documents (architecture, requirements, constraints, API contracts, non-functional requirements).
  3. •Construct a capability map, boundary map, and preliminary dependency graph.

Phase 2: Structural Synthesis

  1. •Propose major phases reflecting delivery epochs.
  2. •Propose milestones as cohesive capability bundles.
  3. •Propose epics as concern-aligned implementation clusters.
  4. •Flag ambiguities that materially alter decomposition.

Phase 3: Subagent-Orchestrated Epic Decomposition

  1. •Partition candidate epics into decomposition batches.
  2. •Dispatch subagents according to the out-of-order strategy above.
  3. •Require each subagent to emit:
    • •task candidates,
    • •subtasks where justified,
    • •acceptance criteria,
    • •dependency assumptions.

Phase 4: Consolidation and Normalization

  1. •Merge subagent outputs into a single proposal.
  2. •Eliminate redundancy and resolve scope collisions.
  3. •Normalize dependency references and sequencing rationale.
  4. •Validate that each task has a falsifiable done-state.

Phase 5: Confirmation Gate

Before any mutation of .tasks/:

  1. •Present the full proposed hierarchy and decomposition.
  2. •Ask targeted clarifying questions where ambiguity remains consequential.
  3. •Obtain explicit user confirmation.

Phase 6: Creation Handoff

After confirmation, proceed with plan-task-compatible creation semantics.

Required Deliverables

  • •Hierarchy overview (phases, milestones, epics) with rationale.
  • •Epic-level decomposition and task lists.
  • •Acceptance criteria per task/subtask.
  • •Explicit assumptions, risks, and unresolved decisions.

Anti-Patterns

  • •Treating narrative prose as implementation certainty.
  • •Linear epic decomposition that forfeits context gains.
  • •Over-fragmentation into low-value pseudo-tasks.
  • •Concealing ambiguity instead of surfacing it.

Reference

For validation heuristics, read references/decomposition-rubric.md.