claude-plan-phase

Generates interface-freeze phase plans with parallel swim lanes from versioned roadmaps.

2|1|Updated Jun 25, 2026
One-click install
npx skills add https://github.com/Consiliency/agent-harness --skill claude-plan-phase-consiliency
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: claude-plan-phase
Source: https://github.com/Consiliency/agent-harness/tree/main/skills-src/claude/claude-plan-phase
Command: npx skills add https://github.com/Consiliency/agent-harness --skill claude-plan-phase-consiliency

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and assets (resource) components.

What problem does it solve? Planning a single phase of a multi-phase software roadmap is error-prone: interfaces drift between parallel workstreams, file ownership overlaps cause merge conflicts, and acceptance criteria end up untestable. This Skill produces a rigorously structured phase-plan document that freezes interfaces upfront and decomposes work into parallel-safe swim lanes with disjoint file ownership. ## Core Features & Use Cases - Architecture-first phase planning: Reads a versioned roadmap (e.g., specs/phase-plans-v1.md), resolves the target phase, and produces a plan document with interface freeze gates, a lane DAG, per-lane test/impl/verify tasks, and testable acceptance criteria. - Multi-agent consensus and external review: Optional --consensus mode runs 2-3 Plan teammates with distinct architectural framings, and --review-external runs Gemini and Codex CLI reviews of the draft plan. - Mechanical plan validation: Ships scripts/validate_plan_doc.py, a stdlib-only linter that checks required headings, frontmatter roadmap SHA-256, DAG acyclicity, disjoint owned-file globs, test-before-impl ordering, and interface traceability. - Use Case: Given specs/phase-plans-v1.md, run the planner on phase P3 to produce plans/phase-plan-v1-P3.md with frozen contracts and parallel lanes, then hand it to an execution skill that spawns one worktree-isolated teammate per lane. ## Quick Start Ask the AI to plan phase P3 of the roadmap in specs/phase-plans-v1.md with architectural consensus enabled.

Frequently Asked Questions about claude-plan-phase

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I plan a single phase of a multi-phase software roadmap?▼

Invoke the planner with a phase alias or heading, such as `/claude-plan-phase P3`. It auto-detects the highest-version `specs/phase-plans-v*.md` roadmap, resolves the phase, and writes `plans/phase-plan-<VERSION>-<PHASE_ALIAS>.md` with interface freezes and swim lanes.

How do I get multi-agent architectural consensus on a phase plan?▼

Pass the `--consensus` flag to launch 2-3 Plan teammates with distinct framings (minimal change, clean architecture, maximum parallelism). The orchestrator synthesizes their proposals over up to two rounds and records any unresolved dissent in the plan's Execution Notes.

Can I review a phase plan with external AI CLIs like Gemini or Codex?▼

Yes, the `--review-external` flag runs Gemini and Codex CLIs in parallel against the drafted plan using the bundled review prompt, writing results to a `_reviews.md` sibling file. Both CLIs must be installed and authenticated.

Why does my phase plan fail validation before execution?▼

The bundled `validate_plan_doc.py` script fails plans with missing required headings, cyclic lane dependencies, overlapping owned-file globs between lanes, impl tasks without a preceding test task, or a roadmap_sha256 that does not match the referenced roadmap file.

When should I not use phase-based swim lane planning?▼

Skip it for single-file or single-concern changes, pure research questions, or specs without phase structure. Those cases are better served by a simple detailed plan or direct exploration, since lane decomposition and interface freezes add overhead without benefit.