tune

---
name: tune
description: Retune existing Claude skills with Marshmallow graph-backed overlays, create aligned copies or starter skills, and rollback overlays with approval. Use when the user runs /marshmallow:tune or asks to personalize, retune, align, or rollback a skill.
license: MIT
compatibility: Designed for Claude Code with Python 3.9+ available locally.
allowed-tools: ["Read", "Write", "Edit", "MultiEdit", "Glob", "Grep", "AskUserQuestion", "Bash(${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py:*)", "Bash(rg:*)"]
---

# Marshmallow Tune

Claude Code expands `${CLAUDE_PLUGIN_ROOT}` in the commands below. In other
plugin hosts, resolve the plugin root as the directory two levels above this
`SKILL.md` and substitute that absolute path before running a command.

Tune skills only when the user wants Marshmallow to change durable skill
behavior. Skill overlays are optional downstream use of Marshmallow recall. The
graph supplies source-backed defaults; the target skill keeps its own procedure.

## Choose Targets

Run:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" scan-skills --project "$PWD"
```

Recommend skills where personal taste, product judgment, writing style, design
direction, or architectural defaults materially change the result. Avoid
deterministic checklist skills unless the user explicitly asks.

## Draft Overlay

Use recall to navigate, then get every selected node in full before it affects
the overlay:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" recall "<skill|topic|label>"
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" get "<selected-node-id>" --kind graph --json
rg -n "<skill|topic|label>" ~/.marshmallow/indexes
rg -n "<skill|topic|label>" ~/.marshmallow/graph
```

Scaffold a valid overlay skeleton with
`"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" new overlay <skill-name>`
(field reference: `${CLAUDE_PLUGIN_ROOT}/references/overlay-template.md`). Keep
the overlay short, source-backed, and scoped to defaults, quality bars,
anti-patterns, and ask-when rules. Do not copy the full graph into a skill. Use
only the 2-5 graph nodes that actually change this skill.

Before drafting, answer four questions:

- What kind of signal is each selected node?
- What exact evidence supports it?
- What should this skill do differently?
- Where should this alignment not apply?

## Preview And Apply

Preview:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" overlay preview \
  --skill "<skill-path>" \
  --overlay "<overlay-path>"
```

Apply only after explicit approval:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" overlay apply \
  --skill "<skill-path>" \
  --overlay "<overlay-path>"
```

For a read-only or plugin-cache skill, do not edit the cached file. Offer a
writable aligned copy instead:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" overlay apply \
  --skill "<source-skill>" \
  --overlay "<overlay-path>" \
  --aligned-copy
```

## Starter Skill

If no existing skill is worth tuning, preview a starter skill:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" starter preview --overlay "<overlay-path>"
```

Apply only after explicit approval:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" starter apply --overlay "<overlay-path>"
```

## Rollback

Preview:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" overlay rollback --skill "<skill-path>"
```

Apply only after explicit approval:

```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/marshmallow.py" overlay rollback --skill "<skill-path>" --approve
```

Rollback restores bytes from the backup and restores or removes the overlay
store according to the backup record beside it.