Concise Planning

Concise Planning

Version: 9.0 | Updated: 01-July-2026 | Architect: Karim Bhalwani | Deps: thinker

Pipeline position: sequence (3 of 4) - brainstorming -> architect -> concise-planning -> implementer. This skill produces a verb-first checklist. It runs AFTER design is approved and BEFORE code is written.

Dependencies

Load the following via read_file before using this skill. Skills marked ★ have disable-model-invocation: true and cannot self-invoke - they must be loaded explicitly.

• ~/.copilot/skills/thinker/SKILL.md ★ - structured reasoning scaffold; confirms scope is well-understood before planning begins

Goal

Turn a user request into a single, actionable plan with atomic steps.

Execution Order

Run all phases in this sequence. Do not skip or reorder.

1. Scope Check - evaluate scope breadth and completeness before writing anything.
2. Generate Plan - produce the plan using the template below.
3. Annotate - label each action item [SEQ] or [PAR].
4. Plan Self-Review - verify all four checks before presenting.
5. Feature Progress Tracker (opt-in) - create only if .copilot/state/ exists and plan is approved.

Workflow

1. Scan Context

• Read README.md, docs, and relevant code files.
• Identify constraints (language, frameworks, tests).

2. Minimal Interaction

• Ask at most 1–2 questions and only if truly blocking.
• Make reasonable assumptions for non-blocking unknowns.

3. Generate Plan

Use the following structure:

• Approach: 1-3 sentences on what and why.
• Scope: Bullet points for "In" and "Out".
• Action Items: A list of 6-10 atomic, ordered tasks (Verb-first). The plan template below shows the minimum 6-step structure; expand up to 10 steps as needed.
• Validation: At least one item for testing.

Plan Template

# Plan

<High-level approach>

## Scope

- In:
- Out:

## Action Items

[ ] <Step 1: Discovery>
[ ] <Step 2: Implementation>
[ ] <Step 3: Implementation>
[ ] <Step 4: Implementation>
[ ] <Step 5: Validation/Testing>
[ ] <Step 6: Rollout/Commit>

## Open Questions

- <Question 1 (max 3)>

Checklist Guidelines

• Atomic: Each step should be a single logical unit of work.
• Verb-first: "Add...", "Refactor...", "Verify...".
• Concrete: Name specific files or modules when possible.

No Placeholders - Iron Law

Every action item must be immediately executable by the implementer without further clarification.

The following patterns are forbidden anywhere in a plan:

Enforcement: Before finalizing the plan, scan every action item for these patterns. If any are found, the plan is not ready to hand off.

Scope Check (Run Before Writing the Plan)

Before generating action items, answer these two questions:

1. Does this plan span multiple independent subsystems? If yes, flag it: "This work spans N subsystems. Recommend decomposing into N sub-plans for clarity." Proceed only if the user confirms a single combined plan.
2. Is the scope well-understood? If key unknowns exist (missing spec, no schema, ambiguous acceptance criteria), surface them as Open Questions before writing steps. Do not write speculative steps.

If the request is too underspecified to produce any action items even after surfacing 1-2 questions, output only the Open Questions section and a single line: "Planning blocked: resolve the above questions before a plan can be generated." Do not produce a partial plan.

Plan Self-Review (Run After Writing the Plan)

Before handing off or presenting the plan, verify:

1. Spec coverage - Does every requirement in the spec/request map to at least one action item?
2. Placeholder scan - Re-read every action item against the No Placeholders table above.
3. Type consistency - Do the deliverables of step N match the inputs expected by step N+1?
4. Validation present - Is there at least one testing/verification step?

If any check fails, fix the plan before presenting it.

Parallel Decomposition

When generating action items, classify each step's dependency type:

• [SEQ]: Must wait for the previous step to complete (strict dependency).
• [PAR]: Can execute independently alongside other [PAR] items.

When to Parallelize

• If 3+ items are [PAR], add a note directly after the Action Items section: "Parallel execution recommended: steps [list] can run concurrently via subagents; add an aggregation step at the end."
• Parallelizable tasks (independent research, separate module implementations, multi-file test writing) benefit from decomposition.
• Sequential tasks (multi-step reasoning chains, state-dependent pipelines) MUST NOT be parallelized. Splitting sequential context across agents degrades performance by 39-70%.

Decomposition Rules

1. Identify the dependency graph: which steps produce outputs consumed by later steps?
2. Group independent steps: steps with no shared dependencies can run in parallel.
3. Always include an aggregation step: after parallel work, a single agent reviews and integrates all outputs.
4. Never parallelize tool-heavy work: if a sub-task needs 5+ tool calls, keep it in a single agent to preserve context budget.

Example

## Action Items

[SEQ] 1. Define API schema and data models
[PAR] 2a. Implement user endpoint (depends on: 1)
[PAR] 2b. Implement product endpoint (depends on: 1)
[PAR] 2c. Write shared test fixtures (depends on: 1)
[SEQ] 3. Integration test all endpoints (depends on: 2a, 2b, 2c)
[SEQ] 4. Run linter and type checker

Feature Progress Tracker (Opt-In)

When a .copilot/state/ directory exists in the target project, create a machine-readable progress tracker alongside the plan. This enables agents to update task status as they work, and retrospectives to measure velocity.

When to create: After the plan is finalized and approved. Generate one tracker per feature/plan.

Procedure:

1. Load the feature_progress.json template.
2. Populate feature from the plan title, branch from current Git branch (or leave empty).
3. Map each Action Item to a tasks[] entry: title from the step text, id from step number, depends_on from [SEQ]/[PAR] annotations.
4. Set status to "planning" and compute summary counts.
5. Write to .copilot/state/FEATURE_PROGRESS.json.

Field rules (for any agent updating the tracker):

• Only modify status, started, completed, and notes on individual tasks.
• Recompute summary counts and updated date after every change.
• Valid task statuses: not-started, in-progress, completed, blocked.
• Valid top-level statuses: planning, in-progress, review, completed.
• Never rewrite title, id, or depends_on (owned by the planner).

Skip signal: If .copilot/state/ does not exist and the task is a quick fix or one-off, skip tracker creation. Mention: "Feature tracker skipped (no .copilot/state/ directory)."

Outputs & Deliverables

• Primary Output: A single plan.md following the Plan Template
• Secondary Output: Open questions and assumptions list; FEATURE_PROGRESS.json (if opt-in)
• Success Criteria: Plan has atomic action items and at most 3 blocking questions
• Quality Gate: Plan reviewed and approved by requester or implementer

Definition of Done

• Plan follows the template: Approach, Scope (In/Out), Action Items, Open Questions
• Action items are atomic, verb-first, and reference specific files/modules
• At least one validation/testing step is included
• Open questions limited to 3 or fewer blocking items
• Plan reviewed and approved by requester

Constraints

• Technical Constraints: Do not implement code in this step; produce a plan only.
• Scope Constraints: Plans should be limited to the requested scope; avoid speculative features.
• Governance Constraints: Align plan with project-context.md and SPEC.md if present.

Common Pitfalls

• Over-Planning: Creating 30+ atomic steps when 6-10 suffice. Keep it concise; details emerge during execution.
• Vague Action Items: "Implement feature" isn't actionable. "Add payment processing to checkout flow" is specific.
• Missing Validation Steps: A plan without testing/verification sets up for rework. Always include a validation checkpoint.
• Ignoring Dependencies: Listing steps in random order instead of respecting precedence. Steps must be executable in sequence.
• Scope Creep: Including "nice-to-haves" in the main plan. Use "Out of Scope" section to acknowledge but exclude them.
• Assuming All Context: Not reading existing code/docs leads to redundant or conflicting plans. Always scan context first.

Integration Points

References

Load these to calibrate plan structure and output format:

Reference Documents

• plan_template.md - Reusable planning checklist template. Load at the start of every planning task to ensure the output follows the correct atomic-step format.
• feature-plan.md - Worked feature plan example. Load when unsure of appropriate granularity or when the user needs a concrete example to validate the plan structure.
• feature_progress.json - Feature progress tracker template. Load when creating a progress tracker for a plan (opt-in, when .copilot/state/ exists).

Scripts

• format_checklist.py - Checklist formatter and validator. Converts a plain list of tasks into a structured, verb-led planning checklist with effort estimate, hygiene warnings (non-verb items, overly broad tasks), and Markdown output ready for user review.