template-options-and-hooks
Reason about CCDS v2 options, hooks, dependency-file generation, prompt patches, and custom configuration overlays.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- template-options-and-hooks
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Reason about CCDS v2 options, hooks, dependency-file generation, prompt patches, and custom configuration overlays.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md5.6 KB
- 参考文档references/hook-reference.md5.4 KB
- 参考文档references/options-reference.md6.4 KB
- 参考文档references/troubleshooting.md5.5 KB
- 脚本scripts/summarize_ccds_options.py10.8 KB
03 SKILL.md 正文
Template Options And Hooks
Use this sub-skill when a task depends on Cookiecutter Data Science (CCDS) v2 configuration choices, nested prompts, post-generation hook effects, dependency file contents, or custom configuration overlays.
Route basic ccds invocation, output directories, replay, and checkout usage to
../project-generation-cli/. Route how to use a generated project's Makefile,
source package, docs, or tests after generation to ../generated-project-workflows/.
Reference Map
- Read references/options-reference.md when you need the exact CCDS v2 option names, defaults, choices, nested subfields, valid environment/dependency combinations, or
custom_configsemantics. - Read references/hook-reference.md when you need to predict generated files, package additions, tests/docs pruning, license handling, scaffold removal, prompt monkey patches, or hook ordering.
- Read references/api-reference.md when you need helper API signatures, dependency writer behavior, Python version specifier rules, package metadata, or CLI option surface facts.
- Read references/troubleshooting.md when a configuration fails, generated files do not match expectations, dependency files look wrong, optional CLIs are missing, or a custom overlay changes the project unexpectedly.
- Run scripts/summarize_ccds_options.py when you have CCDS-compatible option/help JSON files and need a deterministic Markdown or JSON summary without importing CCDS, MkDocs, or Cookiecutter.
Operating Procedure
- Identify whether the user is asking about configuration semantics or about project generation. Stay here only for option/hook reasoning; route command mechanics and generated-project usage as described above.
- Normalize the option set against the exact names and choices in the options
reference. Preserve nested
dataset_storageshape and remember thatcustom_configis hook-consumed but not a normal prompt field. - Check the environment-manager/dependency-file pair before predicting output. Invalid pairs often bake until a later hook or Makefile phase fails; reject them early using the compatibility table.
- Predict hook effects in order: dependency package list construction, lint setup selection, tests/docs pruning, dependency-file writing, Python version update, custom overlay, license cleanup, pyproject quote cleanup, and optional scaffold removal.
- For dependency-file questions, use the helper API reference rather than assuming all formats contain the same information. Pixi and Poetry are special cases.
- For failures or surprising output, use the troubleshooting reference to map the symptom to the most likely option, hook phase, optional CLI, or overlay cause.
Safety Notes
- Do not instruct future agents to read the CCDS source checkout to answer these questions. The bundled references are the runtime source of truth for CCDS v2.3.0 behavior covered by this sub-skill.
- Treat
custom_configas powerful and potentially destructive: validate any overlay payload before generation because it can collide with generated paths and may be followed by later hook cleanup. - Do not run mutating Cookiecutter hooks outside a disposable generated project. This sub-skill describes hook behavior; it does not bundle the original hook as a runtime script.
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).