Memory To Skill — Development skill for Claude Code

memory-to-skill

A Claude Code plugin that turns the sections of your sessions that actually *worked* into repeatable skills and slash commands — mark an attempt, and when it succeeds, distill it.

[](LICENSE)  

---

The idea

Some sessions produce a result you'll want again: a migration recipe, a debugging procedure, a deploy sequence that finally worked. Rather than hoping you remember the prompts, you delimit the good part with markers and run `/distill`. Claude re-reads that section as *evidence* — extracting the goal, keeping the working path, converting dead ends into "gotchas", generalizing project-specific details — and writes a proper skill or slash command from it.

Highlights

• Two markdown commands, zero runtime. No hooks, no scripts, no Node.js, no dependencies. Works identically on macOS, Linux, and Windows.
• Mark live or after the fact. /mark drops a marker into the session as you work; the identical [SKILL-START: ] string can be pasted into a saved transcript later when you realize yesterday's session had the good stuff.
• Skill or command output. Procedures become auto-triggering skills (SKILL.md); replayable prompts become parameterized slash commands. Claude picks the right form, or you force one.
• Project scope by default. Output lands in your project's .claude/ directory, so committing it shares the skill with your whole team. A personal flag targets ~/.claude/ instead.
• Distills, never copies. The output is instructions for a fresh Claude — dead ends become warnings, specifics become placeholders, and the verification step that proved success comes along.

Prerequisites

| Requirement | Why | How to check