How do I install the Effort Tuner skill?

Run npx skillmds@latest add claude-hangar/effort-tuner in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Effort Tuner safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Effort Tuner?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Effort Tuner free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Effort Tuner?

claude-hangar (@claude-hangar) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Effort Tuner by claude-hangar — Agent Skill

Name: Effort Tuner
Author: claude-hangar

Effort Tuner

Analyzes the current task and recommends the optimal Claude Code effort level (low | medium | high | xhigh | max) based on complexity heuristics. Helps avoid burning xhigh/max on trivial tasks or shipping undercooked analysis at low/medium. Use when: "effort tuner", "what effort", "effort level", "which effort should I use", "effort recommendation".

claude-hangar 37ba91d 2 files · 2.9 KB Updated Sep 17, 2026

File contents

/effort-tuner

Recommends an effort level for the current task. Runs cheap — uses heuristics, not expensive reasoning — so it can sit in front of expensive agent calls without tax.

Effort ladder (Claude Code v2.1.111+)

Level	Token budget	Use case
`low`	small	Known-good mechanical edits, formatting
`medium`	moderate	Focused bug fix, localized refactor, single-file change
`high`	generous	Multi-file features, code review, architecture sketches
`xhigh`	larger still	Deep analysis, cross-cutting refactor, complex planning
`max`	full budget	Opus-scale reasoning: novel design, security audit, race-condition hunt

Decision heuristics

Score the task on these dimensions and sum:

Breadth — files touched: 1 (+0), 2–5 (+1), 6+ (+2)
Novelty — pattern exists in repo (+0) vs. new pattern (+1) vs. novel domain (+2)
Risk — style/doc (+0), logic (+1), security/data (+2)
Reversibility — local change (+0), PR-visible (+1), deployed/shared (+2)
Dependency depth — self-contained (+0), one module (+1), cross-module (+2)

Mapping

Sum	Recommendation
0–2	`low` or `medium`
3–5	`high`
6–8	`xhigh`
9–10	`max`

Output template

Effort recommendation: <level>
Task signals:
  - Breadth: <N files> (+<score>)
  - Novelty: <note> (+<score>)
  - Risk: <note> (+<score>)
  - Reversibility: <note> (+<score>)
  - Dependency depth: <note> (+<score>)
Total: <sum>/10
Rationale: <1 sentence>
If unsure: prefer one step down; you can always re-run with higher effort.

Anti-patterns flagged

max on trivial — recommend downgrade + explain token cost
low on security — reject and force ≥high
xhigh on single-file edit without dependencies — flag as possibly wasteful

effortLevel in settings.json — the session default
/effort slider (v2.1.111) — interactive switch for the current turn
Per-agent effort: frontmatter — agent-level default

---name: effort-tunerdescription: Analyzes the current task and recommends the optimal Claude Code effort level (low | medium | high | xhigh | max) based on complexity heuristics. Helps avoid burning xhigh/max on trivial tasks or shipping undercooked analysis at low/medium. Use when: "effort tuner", "what effort", "effort level", "which effort should I use", "effort recommendation".---# /effort-tunerRecommends an effort level for the current task. Runs cheap — uses heuristics, not expensive reasoning — so it can sit in front of expensive agent calls without tax.## Effort ladder (Claude Code v2.1.111+)| Level | Token budget | Use case ||-------|-------------|----------|| `low` | small | Known-good mechanical edits, formatting || `medium` | moderate | Focused bug fix, localized refactor, single-file change || `high` | generous | Multi-file features, code review, architecture sketches || `xhigh` | larger still | Deep analysis, cross-cutting refactor, complex planning || `max` | full budget | Opus-scale reasoning: novel design, security audit, race-condition hunt |## Decision heuristicsScore the task on these dimensions and sum:1. **Breadth** — files touched: 1 (+0), 2–5 (+1), 6+ (+2)2. **Novelty** — pattern exists in repo (+0) vs. new pattern (+1) vs. novel domain (+2)3. **Risk** — style/doc (+0), logic (+1), security/data (+2)4. **Reversibility** — local change (+0), PR-visible (+1), deployed/shared (+2)5. **Dependency depth** — self-contained (+0), one module (+1), cross-module (+2)### Mapping| Sum | Recommendation ||-----|----------------|| 0–2 | `low` or `medium` || 3–5 | `high` || 6–8 | `xhigh` || 9–10 | `max` |## Output template```Effort recommendation: <level>Task signals: - Breadth: <N files> (+<score>) - Novelty: <note> (+<score>) - Risk: <note> (+<score>) - Reversibility: <note> (+<score>) - Dependency depth: <note> (+<score>)Total: <sum>/10Rationale: <1 sentence>If unsure: prefer one step down; you can always re-run with higher effort.```## Anti-patterns flagged- **max on trivial** — recommend downgrade + explain token cost- **low on security** — reject and force ≥high- **xhigh on single-file edit without dependencies** — flag as possibly wasteful## Related- `effortLevel` in settings.json — the session default- `/effort` slider (v2.1.111) — interactive switch for the current turn- Per-agent `effort:` frontmatter — agent-level default

Frequently asked questions