SkillHone Synthesis
Turns tool exploration into hard, machine-gradable benchmark Q/A instead of trivia lookups.
Test report
- Verdict
- Tested · Works
- Score
- Tested
- Jul 14, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Jul 18, 2026 · c13e75c
Built a 6-employee toy dataset and compared a naive question ('what's E003's salary?', directly readable) against a skill-guided one ('among Engineering hires before 2022, excluding contractors, who has the second-highest salary?') that forces a real filter+exclude+rank chain and ships with an executable verification snippet — the difference between trivia and an actual eval item was concrete and reproducible.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 9/10
- Docs & honesty 5/5
What SkillHone Synthesis does
A methodology skill for building closed-form, mechanically-verifiable benchmark questions by exploring a tool environment into a reusable evidence graph, then mining and validating hard (multi-hop, exclusion/ranking, cross-source) Q/A pairs from it via a Cartographer/Miner/Validator/Deduper pipeline. Use when building an eval set, writing eval items, or generating verifiable-answer datasets from callable tools; explicitly not for open-ended writing or subjective scoring.
How to install SkillHone Synthesis
git clone https://github.com/Tencent/SkillHone
cd SkillHone
mkdir -p ~/.claude/skills
cp -r skills/skillhone-synthesis ~/.claude/skills/skillhone-synthesis
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger SkillHone Synthesis
-
/skillhone-synthesisTurns tool exploration into hard, machine-gradable benchmark Q/A instead of trivia lookups.
It also activates on plain-language prompts like these:
-
Help me build a benchmark of verifiable Q&A pairs from this API -
Generate hard, closed-form eval questions from these tool calls -
I need eval data with exact-match gradable answers for our agent
Frequently asked questions
- Is the SkillHone Synthesis skill free?
- Yes. The skill itself is free from Tencent/SkillHone. SkillProof publishes the install command and an independent test verdict at no cost.
- Does SkillHone Synthesis work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 14, 2026. Verdict: Tested · Works. Built a 6-employee toy dataset and compared a naive question ('what's E003's salary?', directly readable) against a skill-guided one ('among Engineering hires before 2022, excluding contractors, who has the second-highest salary?') that forces a real filter+exclude+rank chain and ships with an executable verification snippet — the difference between trivia and an actual eval item was concrete and reproducible.
- What is the SkillHone Synthesis SkillProof Score?
- 9.6/10 — installs cleanly 5/5, triggers reliably 5/5, output vs. baseline 9/10, docs & honesty 5/5.
- How do I install SkillHone Synthesis?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use SkillHone Synthesis with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.