SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsbreakdown-tasks

breakdown-tasks

Coding & development

View repository

Updated 10-15 21:43

About this skill

Task Breakdown — Orchestrator

Productivity — multi-agent orchestration. Break architecture into executable tasks; consumers implement one at a time.

Core Question: "Can an engineer pick up any single task and ship it independently?"

Methodology (risk-first, vertical slices, stable IDs, multi-agent rationale), principles, when NOT to use: references/playbook.md [PLAYBOOK].

Capabilities surfaced: decompose, task-breakdown, sprint-plan, work-breakdown, acceptance criteria, task dependencies, implementation order, risk-first ordering.

Critical Gates

Critic-agent verifies ALL before delivery:

  • Every task has exactly ONE acceptance test
  • No task depends on something not yet defined
  • Risky/uncertain work is front-loaded
  • All external config is in Prerequisites, not buried in tasks
  • A junior dev could verify each acceptance criterion
  • No task requires unstated knowledge
  • Tasks are vertical slices (each delivers a testable increment through all layers); horizontal-only needs explicit justification

On fail: critic names the agent; orchestrator re-dispatches with feedback. Max 2 rounds; round >2 → DONE_WITH_CONCERNS.

Before Starting

Apply references/_shared/before-starting-check.md [PLAYBOOK]:

# Action
0 Mode resolve per references/_shared/mode-resolver.md. Default standard. Downgrade to fast (<10 tasks, single feature, no upstream) → Single-Agent Fallback. Escalate to deep for multi-feature roadmaps with cross-feature deps.
1 Read implementation-roadmap/canonical-paths.md if present.
2 Read architecture/system-architecture.md + docs/forsvn/artifacts/meta/specs/*.md + docs/forsvn/artifacts/product/flow/*.md. None + no source in chat → recommend /discover or /architect-system.
3 Read existing docs/forsvn/artifacts/meta/tasks.md — re-run behavior in references/task-format.md §"Re-run behavior".
4 No experience/ read — project-specific.

Pre-Dispatch

Per references/_shared/pre-dispatch-protocol.md [PROCEDURE]. Warm/Cold Start prompts, dimensions, write-back: references/procedures/pre-dispatch.md [PROCEDURE]. Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.

Multi-Agent Architecture

5 agents, 2 layers (decomposer + dependency-mapper parallel → ordering → acceptance → critic). Agent Roster, Execution Layers, 6-step Dispatch Protocol, Routing Rules, Single-Agent Fallback: references/procedures/multi-agent-dispatch.md [PROCEDURE].

Artifact Contract

  • Path: docs/forsvn/artifacts/meta/tasks.md (single file; edited in place; full re-decomposition snapshots prior to tasks.v[N].md)
  • Lifecycle: pipeline
  • Frontmatter: skill, version, date, status, stack (=meta), review_surface (=md), decision_state, review_tool, reviewed_at, reviewer. Schema + Status Index + Shared Context + task-block format: references/task-format.md [PROCEDURE].
  • Required sections: Status Index (source-of-truth for resume), Shared Context (decisions every task references), Tasks (sibling ### blocks, stable IDs).
  • Consumed by: fresh engineering sessions running Resume Protocol from references/execution-protocol.md; orchestrators batching AFK; operator; fresh-eyes.

Execution Hand-off

Planner, not runner. After tasks.md ships, the next consumer owns execution. Do not re-run this skill to implement. Operating manual: references/execution-protocol.md (Resume, Per-Task, Update/Remove/Reopen, Concurrency, Staleness, Coding Rules).

Anti-Patterns

references/anti-patterns.md [ANTI-PATTERN] — re-read before any task that smells off.

Worked Example

references/examples/decompose-walkthrough.md [EXAMPLE] — Todo app, end-to-end Layer 1 + Layer 2 + critic PASS.

Durable Rules (protected)

Completion Status

Every run ends with explicit status:

  • DONE — tasks decomposed, sized, ordered, deps + criteria written; critic PASS round 1 or 2
  • DONE_WITH_CONCERNS — complete but sizing/dep/criteria ambiguity, or critic FAILed round 2
  • BLOCKED — work too large/under-specified; needs scope reduction or further discovery
  • NEEDS_CONTEXT — missing spec/architecture/product-context; recommend discover or architect-system

Next Step

Hand off tasks.md to implementing agent/dev with the Resume Protocol pointer. Do not re-invoke for the same scope unless re-decomposing.

References

  • references/: playbook.md [PLAYBOOK], task-format.md [PROCEDURE], execution-protocol.md, anti-patterns.md [ANTI-PATTERN], {sizing-examples, dependency-patterns, acceptance-criteria}.md
  • references/procedures/{multi-agent-dispatch, pre-dispatch}.md [PROCEDURE]
  • references/examples/decompose-walkthrough.md [EXAMPLE]
  • references/_shared/{before-starting-check, mode-resolver, pre-dispatch-protocol}.md