Tune settings

Skill kyuna0312/context_forge/skills/tune-settings

This skill should be used when the user says "optimize settings.json", "tune settings for low tokens", "settings.json for performance", "disable auto memory", "configure for token savings", "minimal context window settings", or "settings taking too many tokens". Only proposes documented settings keys — never invented ones.From its SKILL.md

Install

npx -y skills add kyuna0312/context_forge --skill tune-settings

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.4 KB, as published. Nobody here has run it

Tune Settings

Read and optimize settings.json (or settings.local.json) for minimal token usage. Show diff before writing.

Step 1: Find Settings Files

Check in order:

./.claude/settings.json (project)
./.claude/settings.local.json (project local)
~/.claude/settings.json (global)
~/.claude/settings.local.json (global local)

Read all found files.

Step 2: Analyze Current Settings

Identify token-costly configuration. Only these keys are real — never propose a key you cannot find in the official settings reference; a fabricated key is silently ignored and the "optimization" is a placebo:

Setting	Cost	Recommendation
`autoMemoryEnabled: true` (default)	Medium	`false` if auto memory isn't used
`disableBundledSkills` unset	Medium	`true` to skip bundled skills at startup
Unused MCP servers in `.mcp.json`	Medium	Add to `disabledMcpjsonServers`
`autoCompactWindow` unset	Low	Tune via `/autocompact` if compaction fires too late
Hook scripts that read files	Medium	Audit hooks
Unused plugins	Low–Medium	`/plugin remove <name>` (not a settings key)

Note: skill bodies lazy-load — only frontmatter descriptions cost tokens every session. The biggest constant cost is usually CLAUDE.md itself; check that first (/context_forge:check-claudemd-size) before touching settings.

Step 3: Apply Token-Saving Config

Recommended low-token settings.json (real keys only):

{
  "autoMemoryEnabled": false,
  "disableBundledSkills": true,
  "autoCompactEnabled": true,
  "disabledMcpjsonServers": ["unused-server-name"]
}

Key settings explained:

autoMemoryEnabled: false — Disables auto memory loading each session
disableBundledSkills: true — Skips Claude Code's bundled skills at startup
autoCompactEnabled: true — Auto-compact instead of erroring at context limit (default, keep it)
disabledMcpjsonServers — Rejects listed .mcp.json servers so their tool schemas never load

Step 4: Show Diff and Confirm

Display before/after diff:

- "autoMemoryEnabled": true,
+ "autoMemoryEnabled": false,
+ "disableBundledSkills": true,

Ask user to confirm scope: project settings or global settings.

Step 5: Write File

On confirmation, write the updated settings file.

Warning: Changing global settings affects all Claude Code sessions.

LTX Schema

Emit structured output as LTX rows when reporting settings analysis results.

@v1:setting|current|recommended|action

Field	Description
`setting`	Settings key (e.g. `autoMemoryEnabled`)
`current`	Current value or `missing` if not set
`recommended`	Recommended value for token savings
`action`	`change`, `keep`, `add`

Example:

@v1:setting|current|recommended|action
autoMemoryEnabled|true|false|change
disableBundledSkills|missing|true|add
autoCompactEnabled|true|true|keep

Additional Resources

references/settings-reference.md — Full settings.json reference with token impact ratings