Hebbs Tune

Retrieval tuning skill for HEBBS: profile the client, generate domain-specific evals, run baseline, tune retrieval parameters, store learnings, and export compiled rules to a markdown file for prompt injection.

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add hebbs-ai/hebbs-skill --skill hebbs-tune
Manual install
git clone --depth 1 https://github.com/hebbs-ai/hebbs-skill.git /tmp/hebbs-skill
mkdir -p ~/.claude/skills
cp -r /tmp/hebbs-skill/tune ~/.claude/skills/hebbs-tune

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# HEBBS Tune: Teach Your Agent to Retrieve Better

This skill turns retrieval from guesswork into a measured, repeatable process. You profile the client, generate evals that match their domain, run baselines, tune parameters, and store what works. The end state is a compiled rules file that loads into every conversation before the first tool call.

---

## When to activate

- After `hebbs init` + `hebbs index` completes on a new vault
- When the user says "tune", "optimize", "improve recall", "why can't it find X"
- When recall results are visibly poor (low scores, missing obvious facts)
- When vault content changes significantly (new batch of files indexed)

---

## Phase 1: Profile the client

**Do not generate evals yet.** Ask the user first. You need to understand what they search for, not what's in the files.

### Questions to ask

Ask these conversationally. Don't dump a questionnaire. Weave them into the conversation based on what you already know.

1. **What's the domain?** Legal contracts, sales calls, engineering docs, research papers, support tickets?
2. **What do you typically search for?** Specific facts, timelines, decisions, comparisons across documents, contradictions?
3. **Who uses the results?** You (the agent) autonomously, or a human reviewing your output?
4. **What does a wrong answer cost?** Compliance risk, lost deal, wasted debugging time, or just mild annoyance?
5. **What's the hardest thing to find right now?** This becomes your first hard eval.
Read the full skill on GitHub

Details

Skill name
hebbs-tune
Path
tune/SKILL.md
Repo updated
30 March 2026

FAQ

What does the Hebbs Tune skill do?

Retrieval tuning skill for HEBBS: profile the client, generate domain-specific evals, run baseline, tune retrieval parameters, store learnings, and export compiled rules to a markdown file for prompt injection.

How do I install the Hebbs Tune skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add hebbs-ai/hebbs-skill --skill hebbs-tune. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Hebbs Tune popular?

It lives in hebbs-ai/hebbs-skill, which has 9 GitHub stars and 0 forks.

Related skills

Hebbs

9

hebbs-ai/hebbs-skill

Cognitive engine for AI agents: index files and memories, retrieve across 4 weighted dimensions (similarity, temporal, importance, frequency), automatic insight consolidation, automatic contradiction detection, with full parameter control on every function.

AI, LLMs & Agents

Lightweight Explorer

9

kadaliao/claude-code-skills-collection

轻量探索模式 — 用最少 token 完成代码搜索和调研。核心理念:只读搜索不需要加载全部工具和项目规则。适用于:(1) 搜索代码库找文件/函数 (2) 调研项目架构 (3) 多文件分析 (4) 回答"代码里哪里有 X"类问题。灵感源自 Claude Code Explore Agent 设计。

AI, LLMs & Agents

Self Rationalization Guard

9

kadaliao/claude-code-skills-collection

自我合理化防护 — 识别和反制 AI Agent 常见的偷懒/逃避/自我欺骗模式。不限于验证,适用于所有任务执行。当 Agent 即将跳过步骤、简化问题、或用"看起来对"替代"运行验证"时触发。灵感源自 Claude Code Verification Agent 的 Rationalization Detection。

AI, LLMs & Agents

Gemini Nano Banana 2

9

tamtom/image-generation-skills

Generate and edit images using Google's Gemini Nano Banana 2 (gemini-3.1-flash-image-preview). Use when the user asks to generate, create, edit, modify, change, alter, or update images using Gemini. Also use when user references an existing image file and asks to modify it in any way. Supports text-to-image generation, image editing, multi-image compositing (up to 14 reference images), style transfer, Google Web Search and Image Search grounding for real-time data, high-resolution output up to 4K, controllable thinking levels, 14 aspect ratios, and advanced text rendering. DO NOT read the imag

AI, LLMs & Agents

Worker Prompt Craft

9

kadaliao/claude-code-skills-collection

Worker/子代理 Prompt 编写指南 — 写出让子代理一次做对的 prompt。解决"子代理看不到上下文"、"指令模糊导致返工"、"委托理解而非综合"的问题。适用于:(1) 使用 sessions_spawn/sessions_send 时 (2) 调度 CC 子任务时 (3) 任何需要给 AI 子代理写指令的场景。灵感源自 Claude Code Coordinator Prompt Guidelines。

AI, LLMs & Agents

Event Modeling

9

noahseger/agent-skills

Use to model a system's behavior as domain events, extract an event model from existing code, design event-driven architectures, or document workflows as timelines of commands, events, and state views

AI, LLMs & Agents

Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.