openspec-plan

OpenSpec Plan

Analysis + reasoning engine for change proposals. Assumes decision made (via /brainstorm or direct request).

Workflow: Analyze → Reason → Generate

Phase	Purpose	Model	Thinking
Analyze	Explore codebase, find affected files	Sonnet	off
Reason	Dependencies, risks, integration points	Sonnet	on
Generate	Write artifacts from analysis	Sonnet	off

Exploration: Consult openspec/project.md → Exploration Strategy section for tools and must-read files.

Philosophy: Consult openspec/project.md → Execution Philosophy section for current mode and principles.

Commands

create

Create change proposal from codebase analysis.

Input: $ARGUMENTS = change-id: Brief description

Workflow:

1. Analyze: Explore codebase for affected files/patterns (Glob, Grep, Read)
2. Reason (thinking on): Identify dependencies, risks, integration points
3. Philosophy check: Read Execution Philosophy mode, align proposal scope with principles
4. Research (optional): Web search for framework docs/patterns if needed
5. Confirm: Present findings to user before writing
6. Generate: Write proposal.md with analysis-driven content

Philosophy alignment: Include ## Execution Philosophy Alignment section in proposal (see reference.md for template).

Output: openspec/changes/{change-id}/proposal.md with Why/What/Impact from analysis

Next step (ALWAYS display after writing proposal.md): ✅ proposal.md written → Next: /openspec-design design {change-id} → Then: /openspec-plan tasks {change-id}

tasks

Generate outcome-centric tasks + test strategy from proposal.

Input: $ARGUMENTS = change-id [--skip-design] [--skip-test]

Sources (read before generating):

1. openspec/changes/{id}/proposal.md (required)
2. openspec/changes/{id}/design.md (required unless --skip-design)

Design gate: If design.md does NOT exist and --skip-design not passed:

• ⛔ STOP — do not generate tasks
• Display: ⚠️ No design.md found. Run /openspec-design design {change-id} first, or pass --skip-design to bypass.
• Wait for user decision

design.md mapping rules (when design.md exists):

• Containers → task sections: Each C4 container maps to a numbered section in tasks.md
• Aggregate boundaries → task granularity: Entities within an aggregate = single task; entities across aggregates = separate tasks
• TT interaction modes → task sequencing: Collaboration mode = sequential tasks (tightly coupled); X-as-a-Service mode = parallelizable tasks (clear interface)

If --skip-design passed without design.md, derive sections from proposal.md (backward compatible).

Rules:

• Tasks are outcomes: "X exists", "Y works", "Z passes"
• NO activity verbs: Design, Implement, Create, Write
• Each task ≤80 chars (TodoWrite compatible)
• Numbered sections with checkboxes
• Add ### GATE N: Description after sections representing logical boundaries
• Gates optional — not every section needs one
• Standard pattern: scaffold → gate → implement → gate → audit → gate → test

Cross-check (only when design.md was consumed — after generating, before presenting output): Print a coverage table mapping design.md components to generated artifacts:

design.md Component	Type	tasks.md	tests.md	Status
{name}	container/aggregate/flow/invariant/interaction	S{n} / —	GATE {n} / —	✅ / ⚠️

Coverage rules:

• Every C4 container → ≥1 task section
• Every aggregate → tasks covering its entities
• Every key flow → ≥1 gate exercising it
• Every invariant (BC Scope) → ≥1 task enforcing it
• Every context interaction (non-Separate Ways) → tasks or test verification

⚠️ gaps = warn, don't block. If gaps found, ask user: "Fix these gaps now or defer?"

tasks.md headers (MUST include both):

• **Implementation**: Always use /openspec-develop {change-id} — never implement tasks ad-hoc.
• **Design**: See design.md for BC scope, invariants, C4 containers, key flows, and ADRs. (only when design.md was consumed)

Next step (ALWAYS display after writing tasks.md/tests.md): ✅ tasks.md written → Next: /openspec-review {change-id} (tech lead gate) → Then: /openspec-develop {change-id}

Output:

1. tasks.md - verifiable outcomes with gates
2. tests.md - verification strategy (unless --skip-test or garage mode + simple change)

tests.md generation:

• Mirrors gates from tasks.md
• For each gated task outcome: concrete verification step + observable expectation
• Tag each step: [auto] (default), [smoke], or [manual] — see reference.md §Test Execution Model
• Quality bar: functional > structural, observable, specific (see reference.md for anti-patterns)
• Mode-aware: garage=recommended, scale/maintenance=required (see reference.md for requirements)
• MUST include in tests.md header: **Execution**: Always run gates via /openspec-test {change-id} {N} — never ad-hoc.
• MUST include in tests.md header (when design.md was consumed): **Design**: See design.md for BC scope, invariants, C4 containers, key flows, and ADRs.

spec

Generate spec with delta headers and real scenarios.

Input: $ARGUMENTS = change-id capability-name

Output: specs/{capability}/spec.md with:

• ## ADDED/MODIFIED/REMOVED Requirements
• ### Requirement: with SHALL/MUST keywords
• #### Scenario: with GIVEN/WHEN/THEN (from codebase understanding)