claude-plan-detailed

Writes an implementable plan for one bounded code change with verification and handoff.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes assets (resource) components.

What problem does it solve? Planning a small, single-concern code change often gets either skipped entirely or buried under heavyweight roadmap tooling. This Skill produces a concrete, immediately implementable plan document for one bounded change — a bug fix, small feature, or targeted refactor — without the overhead of a full phase roadmap pipeline. ## Core Features & Use Cases - Parallel repo reconnaissance: Launches Explore subagents to map existing code, reusable patterns, and hidden coupling before proposing any change. - Explicit change enumeration: Every planned change names the file, entity, action, and reason, with modification-over-creation and scope-discipline rules enforced. - Verification and acceptance criteria: Requires runnable shell/test commands and testable acceptance assertions, plus an explicit documentation-impact decision. - Optional external review: With --review-external, the plan is critiqued by Gemini and Codex CLIs via a shared review script. - Reflection and handoff closeout: Writes a repo-agnostic reflection and a repo-specific handoff so a fresh implementer session can pick up the plan safely. - Use Case: You need to add a refresh-token endpoint to an existing auth module. Invoke the Skill with the task description; it researches the codebase, writes a plan to .consiliency/plans/, commits it, and leaves a validated handoff for the implementing session. ## Quick Start Ask the agent to plan a bounded change, for example: run claude-plan-detailed with the task "fix stale cache eviction in the session store" and it will research the repo and write an implementable plan document.

Frequently Asked Questions about claude-plan-detailed

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

FAQPage Schema
How do I write an implementation plan for a small code change with Claude Code?▼

Invoke claude-plan-detailed with a free-form task description. It researches the repository with Explore subagents, then writes a plan to .consiliency/plans/ containing enumerated changes, verification commands, and acceptance criteria, and commits the artifact.

When should I use a detailed plan instead of a phase roadmap?▼

Use a detailed plan when the change is bounded — a bug fix, small feature, or targeted refactor with obvious blast radius — and one agent carries it end-to-end. Multi-concern work that benefits from parallel execution should use the phase roadmap builder and per-phase planning skills instead.

Can the plan be reviewed by external AI models before implementation?▼

Yes. Passing --review-external runs the plan through Gemini and Codex CLI review using the shared review script and the bundled review prompt. Both CLIs must be installed and the frontier-model cache populated; results are written to a sibling _reviews.md file.

What are the limitations of a bounded detailed plan?▼

The Skill enforces a bounded-plan threshold: if a plan would touch more than about eight source files or require more than three conceptually distinct changes, it recommends splitting the work into multiple plans. It is also not suited for pure research questions or trivial one-step edits.

How does the handoff work for a fresh implementing session?▼

The Skill writes a repo-specific handoff file with from, timestamp, and artifact fields. A fresh implementer must verify the from field matches, the timestamp is within seven days, and every artifact path resolves under the current repository root before acting on it.