Persistent Planning Skills Pack for Claude Code

Never lose context on multi‑step Claude Code projects. The Persistent Planning Skills Pack adds a three‑file planning system that records your task definition, research findings, and session progress. By keeping this information in task_plan.md, findings.md, and progress.md, the pack prevents Claude Code from discarding critical details when a conversation exceeds its token window. The result is a repeatable workflow that works the same way production AI agents do, letting you move from design to implementation without manual note‑taking.

What You Get

• /plan Skill – Creates and maintains the three planning files. task_plan.md stores phases, goals, constraints and risks; findings.md captures research updates every two tool calls; progress.md logs timestamps, errors and test results. The skill follows a 275‑line structured process that updates automatically as you work.
• /scope Skill – Runs a seven‑dimension complexity analysis (file count, phases, research needs, risk, dependencies, uncertainty, session span). It returns a Simple / Medium / Complex rating and recommends the appropriate planning depth, helping you avoid over‑planning simple tasks or under‑planning large ones.
• /retro Skill – Generates a post‑completion retrospective. It writes memory entries, suggests which findings should be promoted to CLAUDE.md, and archives the three planning files cleanly for future reference.
• Installation assets – Ready‑to‑copy skill files and command definitions. Use the provided cp commands to place them in .claude/skills/ and .claude/commands/.

How to Use

• Copy the skill package: cp -r skills/* .claude/skills/ and the command definitions: cp commands/*.md .claude/commands/.
• Start a new session and invoke /plan. The skill creates task_plan.md, findings.md, and progress.md in the current workspace.
• As you add phases or run tools, the /plan skill updates the files automatically, keeping research discoveries in findings.md and logging each step in progress.md.
• When you need to gauge effort, run /scope. Review the returned complexity rating and follow the suggested planning depth.
• After the task is finished, invoke /retro. The skill writes a concise retrospective, creates memory entries, and optionally moves valuable findings to CLAUDE.md before archiving the three files.

Who This Is For

• Developers building multi‑file features that require hours of incremental work.
• Anyone who has experienced lost context when Claude Code compresses the conversation history.
• Teams that need a reliable hand‑off mechanism between developers during a shared session.
• Architects planning complex refactors or integrations where risk, dependencies and uncertainty must be tracked.