Skill v1.0.1
Automated scan100/100+2 new
version: "1.0.1" name: create-plan description: Create a persistent structured plan (plan.md, phases/, todo.md) when work needs multi-phase or multi-session coordination, durable tracking, or an explicitly requested plan. license: MIT compatibility: opencode metadata: category: planning phase: initial
Skill: Create Plan
What This Skill Does
Creates a complete plan with the artifacts required to execute work in phases. This documentation shall serve Agents and Humans when working in consecutive sessions with the project:
- Plan (
plans/<name>/plan.md) - Problem/context, target outcome, guiding decisions, requirements, DoD, phases overview - Phases (
plans/<name>/phases/phase-N.md) - Scope definition per phase (what/why) - Todo List (
plans/<name>/todo.md) - Trackable items with status - Creates the directory structure for future Handovers
This skill intentionally does not author per-phase implementation plans (plans/<name>/implementation/phase-N-impl.md).
Default next step (second pass): use author-and-verify-implementation-plan to author/verify the per-phase implementation plan against current repo reality.
When to Use
- When work is multi-phase or too large for a single session.
- When the user explicitly requests a persistent plan.
- When durable coordination or tracking is needed across agents, contributors, or sessions.
Do NOT use for a single bounded work package that can be gated with an inline brief containing its task, DoD, constraints, and final verification. Significance or non-triviality alone does not require plans/ persistence.
Execution Model (Recommended)
- Preferred: the primary agent runs this skill and writes artifacts under
plans/<name>/. - Rationale: plans are conversation-anchored (requirements, trade-offs, sequencing, DoD). Moving authorship to a subagent risks losing intent and introducing gaps.
- Use
doc-explorerfor codebase impact/symbol analysis when the plan touches existing code (doc-explorer writes findings todocs/, keeping analysis out of the primary's context). - Optional (edge cases): use
doc-exploreronly to materialize/format a large set of plan files after the primary has finalized content and structure.
Routing Matrix (Who does what)
- Writes:
plans/<name>/plan.md,plans/<name>/phases/**,plans/<name>/todo.md, and creates directories underplans/<name>/. - Does NOT write:
plans/<name>/implementation/**(useauthor-and-verify-implementation-plan). - Primary: owns requirements, scope, DoD, phase breakdown, gating.
- doc-explorer: optional for repo-anchored analysis persisted into
docs/**and/or mechanical plan materialization. - implementer: never used for plan authoring.
Workflow
Step 1: Understand the Objective
Gather requirements from the user using the question tool:
- What is the goal? (feature, bugfix, refactoring, migration)
- What problem or current state motivates the change, and what target outcome should replace it?
- Which user-approved decisions and constraints bind the solution?
- What are the functional requirements?
- What are the non-functional requirements? (performance, compatibility, etc.)
- What is explicitly out of scope?
- What defines "done"? (Definition of Done)
- What testing strategy is expected?
- Are there assumptions that materially bound scope or acceptance?
If the user provided a detailed brief, extract these from the brief and confirm with the question tool.
Step 2: Analyze the Codebase (if applicable)
If the plan involves changes to existing code:
- Use the Task tool with
doc-explorerto analyze the affected modules and symbols (results are written todocs/) - Read existing project documentation (
docs/overview.md, module docs) if available - Identify dependencies and potential risks
- Carry decision-relevant context into the plan, but reference detailed architecture documentation instead of reproducing it.
Step 3: Design the Phase Structure
Determine if phasing is needed:
Single-phase plans (work that otherwise warrants persistence):
- One phase covering the entire scope
- Still create the full directory structure for consistency
- Create this structure only because the work needs persistent coordination or tracking, or the user requested it; a single-phase shape alone is not a reason to create a plan.
Multi-phase plans (complex features):
- Each phase must be completable in a single session
- Phases should have clear boundaries - no phase should depend on "half-done" work from another phase
- Each phase should produce a testable, committable result
- Order phases by dependency (foundational first, then building on top)
Guidelines for phase sizing:
- A phase should represent roughly one focused work session
- If a phase requires reading/modifying more than ~15-20 files, consider splitting
- Each phase should end with passing tests and a clean commit
Step 4: Create the Plan Document
Create plans/<name>/plan.md:
- Concise problem/current-state context and target outcome
- Binding guiding decisions and constraints
- Assumptions only when they materially bound scope or acceptance
- Functional and non-functional requirements
- Scope (in/out)
- Definition of Done
- Testing strategy
- Phases table with titles and brief descriptions
- Risks and open questions
- Initialize the changelog
Step 5: Create Phase Documents
For each phase, create plans/<name>/phases/phase-N.md:
- Phase objective and why the phase exists
- How the phase contributes to the plan goal
- Scope: what this phase includes and explicitly excludes
- Prerequisites (what must be true before starting)
- Deliverables (concrete outputs)
- Acceptance criteria (how to verify the phase is done)
- Dependencies on other phases
Step 6: Create the Implementation Directory (Second Pass Follows)
Create the implementation directory structure (no files yet):
plans/<name>/implementation/
Then, as the default second pass (before execution), author and verify per-phase implementation plans via author-and-verify-implementation-plan.
Step 7: Create the Todo List
Create plans/<name>/todo.md:
- Populate with items from Phase 1 (the starting phase)
- All items start as "Pending"
- Fill in the Phase Context section with links to the phase doc and relevant module docs
- Add an implementation plan link placeholder (to be created by
author-and-verify-implementation-planbefore execution) - Initialize the changelog with the plan creation entry
Step 8: Create the Handover Directory
Create plans/<name>/handovers/ directory with a .gitkeep file.
Step 9: Review with User
Present the plan summary to the user:
- Total phases with brief descriptions
- Key requirements and DoD
- Identified risks
Use the question tool to confirm the plan or gather adjustments.
Rules
- File-based interface: All artifacts go into
plans/<name>/directory structure. The directory name should be lowercase, hyphenated, descriptive. - Phase independence: Each phase must end in a stable state. No phase should leave the codebase broken.
- Phase describes scope: Keep phase docs focused on what/why and acceptance criteria. The per-phase implementation approach (how) is authored later via
author-and-verify-implementation-plan. - Reference, don't duplicate: Implementation plans reference module docs and phase docs. Don't repeat requirements from the plan in each phase.
- Realistic sizing: Phases must be completable in a single session. When in doubt, make phases smaller.
- No built-in explore agent: Do NOT use the built-in
exploresubagent type in this framework. - Use `doc-explorer` for codebase analysis: Delegate deep symbol/dependency analysis via the Task tool. Results are written to
docs/, not returned as text. - Two-pass default for implementation plans: Implementation plans are authored/verified separately via
author-and-verify-implementation-planbefore executing a phase. - Always ask for confirmation: Use the
questiontool to validate requirements, phase structure, and scope with the user before creating artifacts. - Initialize changelog: The plan's changelog should document its creation with the current date.
- Create all directories: Ensure the full directory structure exists:
plans/<name>/,phases/,implementation/,handovers/. - Preserve framing without duplication: Include the problem/current state, target outcome, and binding guiding decisions needed to interpret scope. Reference detailed architecture docs rather than turning the plan into an architecture inventory.
- Material assumptions only: Record an assumption only when it constrains scope, acceptance, or a later decision.
Templates
This skill includes normative templates as bundled files. Only read the templates when processing them. Output MUST follow the template headings and frontmatter keys:
tpl-plan.md- Structure for the plan documenttpl-phase.md- Structure for phase documentstpl-todo.md- Structure for the todo list