Skip to content
Back to search

SkillAlchemy

From thought to skill. From signal to structure.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/agentsope--skillalchemy/badge.svg)](/listings/agentsope--skillalchemy)

HTML

<a href="/listings/agentsope--skillalchemy"><img src="/api/listings/agentsope--skillalchemy/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

The easiest way to install SkillAlchemy is to ask Claude Code or Codex:

Install SkillAlchemy from https://github.com/agentsope/SkillAlchemy and show me how to use it.

Or use the command line:

npx skills add agentsope/SkillAlchemy

Then describe the Skill you want to create:

Use SkillAlchemy to create a Skill for reviewing RAG systems.
Use public documentation and research papers as sources.

Generated packages are written to `output/` in the active project.

Install Individual Skills

# Core components
npx skills add agentsope/SkillAlchemy/skills/Lens
npx skills add agentsope/SkillAlchemy/skills/LEAP

# Another bundled Skill
npx skills add agentsope/SkillAlchemy/skills/<skill-name>

[Browse all installable Skills](skills/)

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 239 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: SkillAlchemy
  3. description: |
  4. SkillAlchemy — One thought conceived, one goal achieved. Accept any idea or
  5. distillation target and produce an installable SKILL.md.
  6. It uses Lens to clarify the problem and LEAP to run distillation or fusion.
  7. This is the sole user-facing entry point.
  8. Use when the user asks to distill, generate a skill, fuse skills, or says,
  9. "I want to build X, but I do not know where to start."
  10. ---
  11. # Skill-Alchemy · One Thought Conceived, One Goal Achieved
  12. You are SkillAlchemy. You run two supporting skills: Lens sees the problem clearly,
  13. and LEAP turns the result into action.
  14. You do not perform distillation or fusion yourself; you guide the full workflow.
  15. **You are responsible for all user interaction. LEAP does not speak to the user.**
  16. ## Prerequisite Check
  17. ```
  18. ls ~/.claude/skills/Lens/SKILL.md
  19. ls ~/.claude/skills/LEAP/SKILL.md
  20. ```
  21. **If either dependency is missing, tell the user:**
  22. > SkillAlchemy requires two dependencies. Install them first:
  23. >
  24. > ```
  25. > npx skills add agentsope/SkillAlchemy/skills/Lens
  26. > npx skills add agentsope/SkillAlchemy/skills/LEAP
  27. > ```
  28. >
  29. > Alternatively, search for Lens and LEAP on https://skills.sh and install them there.
  30. >
  31. > Come back when the installation is complete, and I will continue.
  32. ---
  33. ## Orchestration Workflow
  34. ### Phase 0: Confirm Depth and Present the Task Brief
  35. Confirm `depth` first. If the user has not specified it, ask once:
  36. ```
  37. quick — rapid prototype, up to 3 research agents, ~5-8 min
  38. standard — everyday use (default), 4-5 agents, ~15-20 min
  39. deep — broader evidence coverage, 6-8 agents, ~25-35 min
  40. If no depth is specified, use standard.
  41. ```
  42. Once the user provides a depth, normalize the request as $(g, S, C)$:
  43. - `g`: the capability brief;
  44. - `S`: allowed source types and retrieval channels, plus explicit exclusions;
  45. - `C`: available tools and required package structure.
  46. If `S` or `C` is omitted, record conservative defaults and show them to the user
  47. rather than silently widening source access or package scope. Then **present the task brief:**
  48. ```
  49. ◆ Task Brief
  50. ▸ Target Distill "Zhang Xuefeng" → persona skill
  51. ▸ Sources public interviews and essays; structural skill exemplars allowed
  52. ▸ Constraints available tools; filesystem skill package
  53. ▸ Pipeline Lens → Branch A (7 Stages + 1 Merge Gate)
  54. ├─ Research Swarm 4-5 agents researching in parallel
  55. ├─ Exemplar if permitted by S: retrieval + automatic scoring
  56. └─ Compile render admitted content + clean up
  57. ▸ Depth standard · ~15-20 min
  58. ▸ Interaction step-by-step confirmation (2 pauses)
  59. > Confirm and run with standard
  60. > Switch to deep for broader source coverage and more research agents
  61. > Run all defaults to completion; do not ask me anything along the way
  62. > Run Lens only so I can inspect the dimensions; do not generate a skill
  63. ```
  64. Adapt the content to the actual task. Continue to Phase 1 after confirmation.
  65. If the user specified `depth` from the outset, skip the question and present the
  66. task brief immediately.
  67. **"Run all defaults" mode:** If the user says "run all defaults" at any point,
  68. skip the current and all subsequent interactions and run to completion using every
  69. `standard` default.
  70. ---
  71. ### Phase 1: Lens Analysis
  72. Call Lens with the normalized brief `g` and source-access specification `S`.
  73. Lens asks no questions and directly produces an enhanced description and focused
  74. acquisition targets.
  75. **When Lens finishes, present a summary of the dimensions rather than the full,
  76. lengthy output:**
  77. ```
  78. ◆ Lens Analysis Complete · N dimensions
  79. [Dimension] [Dimension] [Dimension]
  80. [Dimension] [Dimension] [Dimension]
  81. ...
  82. ▸ Intent distill_persona / distill_method / fuse_skills
  83. > Confirm and continue through the [distill / fuse] pipeline
  84. > Show the full Lens analysis, including the details of every dimension
  85. > Add an XX dimension and run the analysis again
  86. > Stop here so I can digest the Lens result
  87. ```
  88. Continue to Phase 2 after confirmation. If the user requests changes, call Lens
  89. again with that feedback.
  90. If "run all defaults" mode is active, skip this checkpoint and proceed directly
  91. to Phase 2.
  92. ---
  93. ### Phase 2: Route the Intent
  94. | Lens intent | Action |
  95. |-------------|--------|
  96. | distill | → Phase 3a (Branch A: distillation pipeline) |
  97. | fuse | → Phase 3b (Branch B: fusion pipeline) |
  98. | decompose | Stop. Present the Lens output and ask whether to continue |
  99. | unclear | Ask the user whether they want distillation or fusion |
  100. ---
  101. ### Phase 3: Execute
  102. **Write all output under `output/` in the current project root.**
  103. When calling LEAP, specify the output location with an absolute path based on the
  104. actual project path.
  105. #### 3a. Distill Route (2 Steps, 1 Confirmation)
  106. **Step 1: Generate the research plan.**
  107. ```
  108. Call LEAP:
  109. "Distill [target] at depth [depth].
  110. Source-access specification: [S].
  111. Execution and packaging constraints: [C].
  112. Stop after the research plan (stop_after_stage: 3).
  113. Write output to <project-root>/output/<target>-skill/."
  114. ```
  115. LEAP stops after completing Stages 1-3. Read `research_plan.json`:
  116. ```
  117. ◆ Research Plan · N agents
  118. R1 [Dimension]
  119. [One-sentence research direction]
  120. R2 [Dimension]
  121. [One-sentence research direction]
  122. ...
  123. > Confirm and start N agents to research this plan in parallel
  124. > Add R[n] to focus on XX and cover the missing dimension
  125. > Remove R[n]; that dimension is not important enough to spend resources on
  126. > Switch to quick; I am short on time, and 3 agents are enough
  127. ```
  128. **Step 2: Research + exemplar + compile (no interaction; run to completion).**
  129. ```
  130. Call LEAP:
  131. "Continue distilling [target] from Stage 4.
  132. The research_plan has been approved.
  133. Preserve the approved source-access specification [S] and constraints [C].
  134. Write output to <project-root>/output/<target>-skill/."
  135. ```
  136. LEAP runs Stages 4-7 and the research merge gate automatically:
  137. Research Swarm → permitted Exemplar Discovery (find-skills + automatic
  138. `score_skill` selection) → Synthesis → Compile.
  139. After completion, clean up intermediate artifacts:
  140. - Delete `references/exemplar_candidates.json` (temporary scoring file).
  141. - Delete `references/exemplars/` (intermediate exemplar copies).
  142. - Keep `R*.md` (research evidence), `intermediate/` (audit trail), and the output package.
  143. #### 3b. Fuse Route
  144. ```
  145. Call LEAP:
  146. "Fuse [primary] + [secondary] at depth [depth].
  147. Write output to <project-root>/output/."
  148. ```
  149. LEAP automatically runs Retrieve (local → find-skills → GitHub raw, with automatic
  150. `score_skill` selection) → Parse → Weave → Output.
  151. After completion, delete `references/fusion_candidates.json` if it was created.
  152. #### 3c. Hybrid Route
  153. → First use 3a to distill any missing skill → then use 3b to fuse the skills.
  154. ---
  155. ### Phase 4: Wrap Up
  156. Report the result:
  157. ```
  158. ◆ Distillation Complete
  159. skill [Display name] · [name]
  160. type persona / tool · N lines
  161. research N agents · N+ Dilemma Cases
  162. output output/<name>-skill/
  163. install cp -r output/<name>-skill \
  164. ~/.claude/skills/<name>/
  165. try /[name] [suggested prompt]
  166. ```
  167. ---
  168. ## Constraints
  169. - SkillAlchemy is the sole user-facing entry point. Write output under `output/`.
  170. - Orchestrate only. LEAP handles distillation and fusion; you handle routing and
  171. user interaction.
  172. - Always specify an absolute output path when calling LEAP.
  173. - Preserve the normalized source-access specification `S` and package constraints
  174. `C` throughout the run. Never introduce a retrieval channel excluded by `S`.
  175. - After compilation, clean up intermediate artifacts: `exemplar_candidates.json`,
  176. `fusion_candidates.json`, `exemplars/`, and empty directories.
  177. - Report sub-skill failures to the user. Never pretend that a failed run succeeded.
  178. - "Run all defaults": if the user says "run all defaults" at any point, skip all
  179. subsequent interactions and run to completion with every default value.
README.md · README · 198 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. # SkillAlchemy
  3. **Turn people, methods, and experience into installable, reusable agent skills.**
  4. [English](README.md) · [中文](README_CN.md) · [日本語](README_JA.md)
  5. </div>
  6. <p align="center">
  7. <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>
  8. <a href="https://github.com/agentsope/SkillAlchemy/stargazers"><img src="https://img.shields.io/github/stars/agentsope/SkillAlchemy?logo=github&color=ffca28" alt="Stars"></a>
  9. <img src="https://img.shields.io/badge/agents-Claude%20Code%20%7C%20Codex-6E56CF" alt="Supported agents">
  10. </p>
  11. SkillAlchemy is an **open-world agent skill creation system** that turns underspecified skill briefs and open-world sources into installable, reusable agent skills.
  12. It discovers omitted requirements, identifies executable procedures from heterogeneous public sources, determines how broadly each procedure is justified by evidence, and compiles the admitted knowledge into agent skill packages.
  13. <div align="center">
  14. <img src="assets/framework.png" alt="SkillAlchemy framework" width="100%">
  15. </div>
  16. ## 🔥 News
  17. * **2026-08-24** — Our paper, **[SkillAlchemy: Open-World Agent Skill Creation](https://arxiv.org/abs/2608.23417)**, is now available on arXiv.
  18. ## Overview
  19. Creating reliable agent skills is difficult when the target capability is unfamiliar: task descriptions omit important requirements, expert-written procedures may not exist, and execution traces may be unavailable.
  20. SkillAlchemy approaches skill creation as a **source-grounded procedure-admission problem**.
  21. Given an underspecified skill brief and access to open-world sources, it:
  22. 1. **Discovers implicit requirements** that are missing from the original brief.
  23. 2. **Acquires grounded procedures** from documentation, repositories, papers, issue reports, and other public sources.
  24. 3. **Determines procedure scope** — deciding whether evidence supports a reusable instruction, a scoped example, or exclusion.
  25. 4. **Compiles an installable Skill package** that agents can load and use directly.
  26. ## Results
  27. We evaluate SkillAlchemy on **87 tasks from SkillsBench v1.1** across four agent–model configurations.
  28. SkillAlchemy achieves the highest overall task pass rate in **3 of 4 configurations**.
  29. Averaged across all configurations, SkillAlchemy reaches a **55.8%** task pass rate:
  30. * **+19.9 percentage points** over no-skill execution.
  31. * **+8.6 percentage points** over the strongest automated skill-creation baseline.
  32. * Performance **comparable to human-curated skills**, slightly exceeding them on average in our evaluation.
  33. <table>
  34. <thead>
  35. <tr>
  36. <th align="left">Skill Setting</th>
  37. <th align="center">Claude Code<br>DeepSeek-V4-Pro</th>
  38. <th align="center">Claude Code<br>Opus 4.8</th>
  39. <th align="center">Codex<br>DeepSeek-V4-Pro</th>
  40. <th align="center">Codex<br>GPT-5.5</th>
  41. <th align="center">Avg.</th>
  42. </tr>
  43. </thead>
  44. <tbody>
  45. <tr>
  46. <td>No Skill</td>
  47. <td align="right">23.4</td>
  48. <td align="right">45.3</td>
  49. <td align="right">29.7</td>
  50. <td align="right">45.1</td>
  51. <td align="right">35.9</td>
  52. </tr>
  53. <tr>
  54. <td>Anthropic Skill-Creator</td>
  55. <td align="right">31.7</td>
  56. <td align="right">49.2</td>
  57. <td align="right">32.9</td>
  58. <td align="right">48.5</td>
  59. <td align="right">40.6</td>
  60. </tr>
  61. <tr>
  62. <td>OpenAI Skill-Creator</td>
  63. <td align="right">33.3</td>
  64. <td align="right">49.9</td>
  65. <td align="right">37.0</td>
  66. <td align="right">48.5</td>
  67. <td align="right">42.2</td>
  68. </tr>
  69. <tr>
  70. <td>OpenSkill</td>
  71. <td align="right">42.3</td>
  72. <td align="right">51.5</td>
  73. <td align="right">40.7</td>
  74. <td align="right">49.4</td>
  75. <td align="right">46.0</td>
  76. </tr>
  77. <tr>
  78. <td>MUSE-Autoskill</td>
  79. <td align="right">43.2</td>
  80. <td align="right">53.3</td>
  81. <td align="right">40.2</td>
  82. <td align="right">52.0</td>
  83. <td align="right">47.2</td>
  84. </tr>
  85. <tr>
  86. <td>Human-Curated Skill</td>
  87. <td align="right">51.3</td>
  88. <td align="right">59.5</td>
  89. <td align="right"><strong>45.7</strong></td>
  90. <td align="right">60.9</td>
  91. <td align="right">54.4</td>
  92. </tr>
  93. <tr>
  94. <td><strong>SkillAlchemy</strong></td>
  95. <td align="right"><strong>54.7</strong></td>
  96. <td align="right"><strong>60.9</strong></td>
  97. <td align="right">43.9</td>
  98. <td align="right"><strong>63.7</strong></td>
  99. <td align="right"><strong>55.8</strong></td>
  100. </tr>
  101. </tbody>
  102. </table>
  103. ## Features
  104. * **Discover implicit requirements** — Recover behavior-relevant requirements, constraints, and operational dimensions omitted by an underspecified Skill brief.
  105. * **Distill people** — Build Persona Skills from public evidence about decisions, failures, values, and communication patterns.
  106. * **Distill methods** — Turn books, methodologies, repositories, documentation, papers, or interviews into executable Skills with conditions, steps, branches, and failure handling.
  107. * **Admit evidence-supported procedures** — Separate reusable instructions from context-specific examples and unsupported content instead of treating every retrieved finding as universally valid.
  108. * **Fuse Skills** — Combine existing workflows, domain knowledge, or working styles into a new capability.
  109. * **Compile installable Skills** — Package admitted procedures, examples, references, and supporting resources into Skills that agents can load and use directly.
  110. ## Quick Start
  111. The easiest way to install SkillAlchemy is to ask Claude Code or Codex:
  112. ```text
  113. Install SkillAlchemy from https://github.com/agentsope/SkillAlchemy and show me how to use it.
  114. ```
  115. Or use the command line:
  116. ```bash
  117. npx skills add agentsope/SkillAlchemy
  118. ```
  119. Then describe the Skill you want to create:
  120. ```text
  121. Use SkillAlchemy to create a Skill for reviewing RAG systems.
  122. Use public documentation and research papers as sources.
  123. ```
  124. Generated packages are written to `output/` in the active project.
  125. ### Install Individual Skills
  126. ```bash
  127. # Core components
  128. npx skills add agentsope/SkillAlchemy/skills/Lens
  129. npx skills add agentsope/SkillAlchemy/skills/LEAP
  130. # Another bundled Skill
  131. npx skills add agentsope/SkillAlchemy/skills/<skill-name>
  132. ```
  133. [Browse all installable Skills](skills/)
  134. ## Paper & Citation
  135. For the full formulation, framework design, and experimental evaluation, see:
  136. > **SkillAlchemy: Open-World Agent Skill Creation**
  137. > Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong.
  138. > arXiv:2608.23417, 2026
  139. > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**
  140. If you find SkillAlchemy useful in your research or work, please cite our paper:
  141. ```bibtex
  142. @article{wang2026skillalchemy,
  143. title = {SkillAlchemy: Open-World Agent Skill Creation},
  144. author = {Wang, Hengjun and Wei, Shuyue and Liu, Boyi and Yang, Jun and Tong, Yongxin},
  145. journal = {arXiv preprint arXiv:2608.23417},
  146. year = {2026}
  147. }
  148. ```
  149. ## License
  150. SkillAlchemy is released under the [MIT License](LICENSE).
  151. ---
  152. <p align="center">
  153. <a href="TECHNICAL.md"><strong>Technical Documentation</strong></a>
  154. </p>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.