skill-authoring
Creates and updates canonical Agent Skills in the Scientific Agent Skills repository while preserving scope, frontmatter, progressive disclosure, versioning, tests, catalog entries, and diagram policy. Use for new skills, skill revisions, metadata fixes, references/scripts/assets placement, or contribution checklists.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- skill-authoring
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Creates and updates canonical Agent Skills in the Scientific Agent Skills repository while preserving scope, frontmatter, progressive disclosure, versioning, tests, catalog entries, and diagram policy. Use for new skills, skill revisions, metadata fixes, references/scripts/assets placement, or contribution checklists.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/authoring-workflows.md2.2 KB
- 参考文档references/catalog-and-diagrams.md1.8 KB
- 参考文档references/frontmatter-and-layout.md2.0 KB
- 参考文档references/troubleshooting.md1.9 KB
- 脚本scripts/audit_skill_frontmatter.py6.8 KB
03 SKILL.md 正文
Skill Authoring
Use this route when the task changes a canonical skill under skills/<name>/.
Before editing
- Read the existing
SKILL.mdand every linked local reference/script that the change may affect. - Read
AGENTS.mdand the relevantCONTRIBUTING.mdsection. - Decide whether the request is a new skill, an update, or a scope correction. Do not create a broad orchestrator skill merely because several neighboring skills exist.
- Use current upstream package/platform documentation for API claims; keep untested behavior explicitly illustrative.
Read references/authoring-workflows.md for the new/update checklists and references/frontmatter-and-layout.md for the exact structural contract. Use scripts/audit_skill_frontmatter.py for a fast local preflight before the repository validator.
New skill route
- Create
skills/<name>/with a matching lowercase-hyphenSKILL.md. - Put only runtime content in the skill directory:
SKILL.md,references/,scripts/, and usefulassets/. - If the skill ships scripts, create tests under
tests/<name>/and add a matching entry totests/skill-requirements.toml. - Keep the root document router-like; move long API tables, examples, data schemas, and troubleshooting matrices into references.
- Update the public catalog and any required generated diagram only after the content is final. Read
references/catalog-and-diagrams.mdbecause this checkout may not contain the diagram generator named by repository policy.
Existing skill update route
- Make the smallest useful change and preserve unrelated behavior.
- Bump the quoted
metadata.versionin the same change. - Re-run touched examples/scripts and the skill's own test suite.
- Re-check local links, line count, optional dependency claims, credentials, and safety boundaries.
- If the workflow or references changed materially, regenerate the diagram when the required tool is available; a typo/link/version-only change does not require an image refresh.
Exit handoff
Before handing off, report the skill directory, files changed, evidence consulted, intentional omissions, version bump, test/validation commands to run, catalog/diagram status, and any upstream or environment uncertainty. Route execution to validation-testing, then route scanner review to security-scanning.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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).