workflows-plan

SkillProductivity

Transform spec (with frontmatter claims) into an implementation plan with task decomposition and sub-agent assignment. Stage 3 of the development workflow.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the workflows-plan skill

What this skill tells your AI

The instructions your AI receives, as published by kastalien-research/thoughtbox in .agents/skills/workflows-plan/SKILL.md and read by ahel’s review.

Create an implementation plan for: $ARGUMENTS

Purpose

You are executing Stage 3 (Planning) of the development workflow. The spec with frontmatter claims exists from Stage 2. Your job is to decompose the work into sub-agent-sized tasks, identify dependencies, and produce a plan file that Stage 4 can execute.

Pre-Conditions

Before starting, verify:

  1. .workflow/state.json exists and currentStage is "planning"
  2. A spec exists (check stages.dev-docs.artifacts.spec in state)
  3. Spec frontmatter includes at least one claim (check stages.dev-docs.artifacts.spec in state)

If pre-conditions are not met, report what's missing and halt.

Process

Step 1: Gather Context

Read the spec (frontmatter + body) from Stage 2, then investigate the codebase:

  1. Read the spec: Understand what needs to be built
  2. Read spec claims: Understand falsifiable claims and required evidence
  3. Map the affected code: Use Glob/Grep to find files that will need changes
  4. Check for prior learnings: Search agent memory and compound learnings for relevant patterns
  5. Check existing tests: Identify test files that will need updates

Step 2: Decompose into Tasks

Break the implementation into discrete, sub-agent-sized units of work. Each task must be:

  • Atomic: One task = one logical change = one commit
  • Testable: Each task has clear acceptance criteria
  • Ordered: Dependencies between tasks are explicit

For each task, define:

### Task N: <title>

**Files**: [list of files to create/modify]
**Depends on**: [task numbers, or "none"]
**Acceptance criteria**:
- [ ] [specific, verifiable condition]
- [ ] [specific, verifiable condition]

**Notes**: [implementation hints, gotchas from codebase investigation]

Step 3: Identify Risks

For each task, flag:

  • Files with high churn or complexity
  • Areas where the spec's assumptions may not hold
  • External dependencies or integration points
  • Tests that may need significant rework

Step 4: Write the Plan File

Write the plan to .workflow/plan.md:

# Implementation Plan: <title>

**Workflow**: <id>
**Spec**: <spec path>
**ADR**: <adr path>
**Branch**: <branch>
**Created**: <ISO timestamp>

## Overview

[1-2 paragraph summary of what's being built and the approach]

## Task Breakdown

[Task definitions from Step 2]

## Dependency Graph

[Simple text diagram showing task ordering]

## Risks

[From Step 3]

## Estimated Scope

- Tasks: N
- Files affected: N
- New files: N
- Tests to write: N

Step 5: Get User Approval

Present the plan summary and ask the user to approve:

PLAN READY FOR REVIEW
======================

<title>
Tasks: N | Files: N | Risks: N flagged

[task list with 1-line summaries]

Approve this plan? (The conductor will dispatch /workflows-work to execute it)

Wait for user approval before proceeding.

Step 6: Record and Handoff

After user approves:

  1. Update workflow state (.workflow/state.json):

    • Set stages.planning.status to "completed"
    • Set stages.planning.completedAt to current ISO timestamp
    • Set stages.planning.artifacts.plan to .workflow/plan.md
    • Set currentStage to "implementation"
    • Update updatedAt
  2. Present the handoff:

    PLANNING COMPLETE
    ==================
    
    Plan: .workflow/plan.md
    Tasks: N
    Dependencies: [summary]
    
    Next: Stage 4 - Implementation (/workflows-work)
    

Anti-Patterns

  • Do NOT start implementing during planning — that's Stage 4's job
  • Do NOT create tasks smaller than one meaningful commit
  • Do NOT create tasks larger than one sub-agent can handle in a single session
  • Do NOT skip codebase investigation — plans based on assumptions fail
  • Do NOT ignore the ADR's hypotheses — the plan must be designed to test them

Signals

GitHub stars
64
Forks
20
Last commit
Jul 2026
Advanced
Catalog kind
skill
Gateway key
workflows-plan
Source
github.com/kastalien-research/thoughtbox