research-skill-creator
Create, improve and test agent skills for research workflows (paper writing, lab protocols, analysis pipelines, domain databases) that meet the Agent Skills specification and this repository's quality and security bar. Use when turning a repeated research task into a reusable skill, contributing a new skill to Research Agent Skills, rewriting an existing skill, writing trigger-accurate descriptions, or adding evals. Works alongside Anthropic's official skill-creator.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- research-skill-creator
- license
- MIT
- metadata
- { "tags": "skill authoring, agent skills, SKILL.md, evals, contributing", "author": "Kalaris Labs (Sayan Chowdhury)", "version": "1.0", "category": "research-automation", "maintainer": "Kalaris Labs" }
- description
- Create, improve and test agent skills for research workflows (paper writing, lab protocols, analysis pipelines, domain databases) that meet the Agent Skills specification and this repository's quality and security bar. Use when turning a repeated research task into a reusable skill, contributing a new skill to Research Agent Skills, rewriting an existing skill, writing trigger-accurate descriptions, or adding evals. Works alongside Anthropic's official skill-creator.
- compatibility
- Python 3.9+ for the scaffold script. Repository validators need uv. Anthropic's skill-creator (optional) provides the eval loop.
02 资源文件 1
- 脚本scripts/new_skill.py6.1 KB
03 SKILL.md 正文
Research Skill Creator
A good research skill packages procedural knowledge an expert would otherwise have to explain every time: the order of steps, the checks that catch mistakes, the venue- or instrument-specific rules. It is not a textbook chapter and not a wrapper around one API call.
0. Install the official tooling (once)
Anthropic's skill-creator supplies the eval loop, description optimizer and
packager. In Claude Code:
/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills
Use it to run evals (run_eval.py, run_loop.py) and to tune the description
(improve_description.py). This skill adds the research-specific conventions.
1. Decide whether it should be a skill
Create a skill when all hold:
- The task recurs across projects or people (not a one-off script).
- Doing it well requires non-obvious steps, checks or domain rules.
- An agent would plausibly get it wrong without guidance (test this: ask without the skill first).
Otherwise write a note or a script instead. Check catalog/skills.json for overlap and
extend an existing skill rather than creating a near-duplicate.
2. Scaffold
python skills/research-skill-creator/scripts/new_skill.py rebuttal-letter \
--category research-writing \
--description "Draft point-by-point responses to peer reviewers. Use when ..." \
--scripts --references
This writes skills/<name>/SKILL.md with spec-compliant frontmatter and evals/<name>/evals.json.
3. Write the description (it decides whether the skill is ever used)
- Say what it does and when to use it, with the words users actually type: "journal cover letter", "response to reviewers", "PRISMA flow diagram".
- Name concrete artifacts, tools and venues ("BibTeX", "Nature", "pgvector").
- At most 1024 characters, third person, no angle brackets, no marketing adjectives.
- Add "Use when ..." clauses for the 3-6 most common triggers, plus distinguishing cues vs. sibling skills (e.g. "for systems venues use systems-paper-writing").
4. Write the body
| Principle | Practice |
|---|---|
| Procedure over prose | Numbered workflows, decision tables, checklists |
| Progressive disclosure | SKILL.md under 500 lines; long material in references/*.md, linked with one-line "read when" hints |
| Deterministic work in code | Parsing, validation and conversion go in scripts/, which the agent runs rather than re-implements |
| Verify-first for facts that change | Venue limits, API versions, guidelines: tell the agent to fetch the official source and cite it |
| Honesty rules | Explicit "never fabricate data, citations or results; say when unverifiable" |
| Handoffs | End with related skills and when to switch |
5. Scripts: the repository contract
- Python 3.9+ standard library unless a dependency is essential. Declare it in
compatibility. argparseCLI with--help, a docstring with usage, exit codes0ok,1findings,2bad input.- UTF-8 console setup (the scaffold includes it) so Windows terminals don't crash.
- Network calls only to documented APIs. Every new domain must be added to
security/allowed-domains.txtwith a reviewer's approval. Set a descriptiveUser-Agent. - Never read credentials beyond the env var the user configures for that service. Never
exfiltrate environment data. No
eval/execof fetched content. Never pipe downloaded scripts into a shell.
6. Test
uv run tools/validate.py skills/<name>: spec + repo policy.uv run tools/lint_injection.py skills/<name>: prompt-injection and payload lint.- Fill
evals/<name>/evals.jsonwith 3-10 realistic prompts, including near-misses that should not trigger, and run them with skill-creator's eval loop. Compare with/without the skill. - Run each script on real sample inputs, including malformed ones, on Windows and Unix if possible.
uv run tools/build_catalog.pyto regenerate the marketplace, catalog and README tables.
7. Rewriting an adapted skill
Skills listed in third_party/upstream-manifest.json with rewrite_status: imported
are adaptations. A rewrite must be written from scratch (new structure, new
wording, re-verified facts), not paraphrased line by line. When it merges, set
rewrite_status: original for that skill in the manifest. The notices generator then
drops it from THIRD_PARTY_NOTICES.md and the importer stops overwriting it.
Related skills
research-agent-skills repository docs (CONTRIBUTING.md), Anthropic skill-creator,
autoskill (detects repeated workflows worth turning into skills).
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.