AGENT SKILL

scientific-agent-skills

Maintains the Scientific Agent Skills repository: adding or updating Agent Skills, validating SKILL.md frontmatter and repo tests, planning isolated skill suites, running or triaging security scans, and keeping catalog/release evidence aligned. Use for K-Dense-AI/scientific-agent-skills checkout tasks, skills-ref validate, tests/_meta, tests/run_all.py, scan_pr_skills.py, scan_skills.py, metadata.version, allowed-tools, openclaw/hermes metadata, docs/skills.md, and scanner findings.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
scientific-agent-skills
license
MIT
metadata
{ "disco-role": "operating" }
description
Maintains the Scientific Agent Skills repository: adding or updating Agent Skills, validating SKILL.md frontmatter and repo tests, planning isolated skill suites, running or triaging security scans, and keeping catalog/release evidence aligned. Use for K-Dense-AI/scientific-agent-skills checkout tasks, skills-ref validate, tests/_meta, tests/run_all.py, scan_pr_skills.py, scan_skills.py, metadata.version, allowed-tools, openclaw/hermes metadata, docs/skills.md, and scanner findings.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Scientific Agent Skills Repository Maintenance

Use this repo skill when the user is editing, reviewing, validating, testing, or preparing a pull request for the K-Dense-AI/scientific-agent-skills repository. It is a maintainer skill for the skill collection itself; it is not the route for using RDKit, Scanpy, Paperclip, or any other scientific workflow skill.

First decision

User intent Read
Add a new skill, update an existing skill, fix frontmatter/metadata, decide whether content belongs in references/, scripts/, or assets/, or update catalog/diagram artifacts sub-skills/skill-authoring/SKILL.md
Choose validation commands, run skills-ref, diagnose tests/_meta, add tests/<skill>/, edit tests/skill-requirements.toml, or use tests/run_all.py --isolated sub-skills/validation-testing/SKILL.md
Run changed-skill or weekly security scans, handle missing SKILL_SCANNER_LLM_API_KEY, interpret scanner findings, or update docs/security-triage.md sub-skills/security-scanning/SKILL.md
Check repository layout, source snapshot, selected evidence, or staleness before refreshing this generated repo skill references/repo-overview.md, references/repo-provenance.md
Diagnose cross-cutting repo failures or confusing policies before touching files references/troubleshooting.md

Repository setup defaults

  1. Work from the repository root of a target checkout.

  2. Prefer the repository's uv workflow for Python 3.13 tooling:

    uv sync --python 3.13
    
  3. Do not install every scientific dependency into the project environment. Individual skills have mutually incompatible package pins; use tests/run_all.py --isolated <skill> for skill-specific suites.

  4. Treat generated scan reports, diagrams, caches, and local environments as outputs, not source evidence.

  5. Before a PR, run the smallest checks that cover the changed skill and then escalate to repo-wide checks when shared contracts, runner logic, or many skills changed.

Common route patterns

  • New skill with scripts: start in skill-authoring, then use validation-testing to add a suite and isolated requirements entry, then use security-scanning to plan the PR scan.
  • Existing skill update: read the current skill first, bump metadata.version, test only changed examples/scripts, and run focused validation before broad checks.
  • CI failure: identify whether it came from skill spec validation, repo-specific metadata rules, tests/_meta, a per-skill suite, or the security scanner; then enter the matching sub-skill.
  • Scanner finding: never fix from the finding text alone. Inspect the cited files, verify whether they exist, compare behavior to the skill's documented service, and check the triage reference for known false-positive classes.

Safety and scope boundaries

  • Do not make medical, scientific, or regulatory claims by editing a skill without evidence from the package/service docs and tests that skill targets.
  • Do not put tests, fixtures, scratch data, or generated reports inside a runtime skills/<name>/ package.
  • Do not add secrets, private URLs, unpublished data, local paths, or environment names to skill files.
  • Do not run networked security scans, upload reports, create releases, or mutate tags unless the user explicitly authorizes that side effect.
  • Avoid using this skill when the task is ordinary scientific analysis with an installed package; route to the package-specific scientific skill instead.

Provenance and refresh

Read references/repo-provenance.md before relying on this generated repo skill for a checkout at a different commit, branch, version, or dirty state. If the repository policy files, CI workflows, test contract, scanner scripts, or catalog layout have changed, refresh this repo skill before using it for high-stakes maintenance.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351