MacHu-GWU/sqlalchemy_upsert_kit-project / .claude/skills/pywf-oss-base
pywf-oss-base
FlaggedShared conventions, references, and scripts of the pywf-oss skills for pywf open-source Python libraries: setup, layout, code style, tests, user docs and the agent skills that hold API and maintenance knowledge, CI and hosting, release, dependencies, and secrets. Load it before ad-hoc work on modules, tests, or docs in such a repo, for a question about these conventions, or when unsure which pywf-oss skill fits a job.
references/00-authoring-references.md
View on GitHub---
description: "How the pywf-oss skills themselves are written: the base and invoker split, reference numbering and backward-only citations, file shape, placeholders, script conventions, and the checks after any change. Read before adding, renaming, or editing any pywf-oss skill, reference, script, or asset."
---
# Authoring the pywf-oss Skills
---
## 1. The family
- **`pywf-oss-base`** holds knowledge only (`references/`, `scripts/`, `assets/`) and is bound to no action. Loaded on its own, it answers convention questions and picks the invoker for a job.
- **Each invoker `pywf-oss-<action>`** does one job. Its body says "Read ../pywf-oss-base/SKILL.md, then:" and lists the base files it needs. Everything after that is its own: the steps, what it asks the user, where it stops for the user's yes, and what it reports.
- **A shared fact lives in exactly one reference.** Invokers cite it and never restate it. If one invoker needs the fact to be different, change the reference.
- **Every skill is model-invocable.** None sets `disable-model-invocation`. A skill that changes an external service asks for the user's yes before each such step instead.
---
## 2. Reference files
- **Name:** `NN-slug.md`, two digits and a kebab-case slug.
- **Order** follows the project lifecycle: authoring (00), secrets and the machine, the toolchain, layout and code, tests, docs, CI and hosting, release, maintenance, and last the skill workflow.
- **Cite backward only.** A reference cites only lower-numbered references. When it needs a fact from later in the lifecycle, move that fact down into the earlier file, or state it without a citation. SKILL.md files may cite any reference.
- **One concern per file.** Split a file when two of its parts can change independently: different tools in one area (Sphinx, its theme, each extension), or structure versus style (where tests live, how they are written). A topic a job needs only sometimes gets its own file, so an invoker can load it on demand.
- **Shape:** YAML frontmatter with only `description`, one H1, then numbered H2 sections separated by `---`.
- **`description`** is one double-quoted string with no double quotes inside, in two parts: what the file holds, then a trigger sentence starting with "Read when" or "Read before". `scripts/update_skill_index.py` copies only the trigger into the base SKILL.md index, which every invoker reads, so keep the trigger short, and specific enough that no two references claim the same question.
- **Citations** look like `` `NN-slug.md` ``, optionally followed by "section N". Prefer the whole file; section numbers move.
---
## 3. Generic content
- Write in English.
- **Use placeholders, never a real project's names,** so the skills work when copied into any repo.
| Placeholder | Meaning |
| :--- | :--- |
| `<package_name>` | The import package directory: snake_case, equal to `[project].name` in `pyproject.toml` |
| `<owner>`, `<repo>` | GitHub owner and repository name. Repositories are named `<package_name>-project`. |
| `module_name`, `subpackage`, `ClassName`, `method_name`, `func_name`, `CONSTANT_NAME`, `attr_name` | Code names in examples |
| `third_party_lib`, `sibling_lib` | An ordinary third-party distribution, and a close-relative distribution |
| `<slug>` | The Read the Docs project slug: the project name with `_` replaced by `-` |
| `<version>`, `<last_tag>` | A version number, and the latest git tag |
| `<cli_name>` | The command name registered in `[project.scripts]` |
| `<skill_name>` | The library's own API skill, `pypi-<package_name>`, written by `pywf-oss-write-pypi-skill` |
| `<maintain_prefix>` | `maintain-<package_name>`, the prefix of the library's own maintenance skill family, written by `pywf-oss-write-maintain-skill` |
| `<dist_name>` | Any PyPI distribution name, in dependency work |
| `<key_path>` | A dotted key path in `$HOME/home_secret.toml` |
| `<run_id>` | A GitHub Actions run id |
| `<path>`, `<source_path>`, `<test_path>`, `<notebook_path>`, `<script_name>` | File system paths and script names in commands |
**Two spellings, on purpose:** a value substituted into a command, path, or config is snake_case in angle brackets (`<package_name>`). A name in example code is a bare identifier (`ClassName`), so the example stays valid Python.
- Name the tools of the template stack freely (mise, uv, ruff, pytest, pytest-cov, Sphinx, furo, docfly, nbsphinx, sphinx-design, sphinx-jinja, home-secret-toml, GitHub Actions, Codecov, Read the Docs): they are the conventions. Name any other library only when a rule is about it, as in "one field per line in pydantic models".
- For a working demonstration, point to the sample pages in `docs/source/` of a template repo. Do not copy a project's code into a reference.
- **Approval wording:** "the user's yes", and "one yes covers one step".
---
## 4. Scripts and assets
- Every script under `scripts/` is a stdlib-only command line tool:
- `_main()` holds the logic, and `main(argv)` only parses arguments;
- `--arg_name` flags only, no positional arguments;
- exit code 0 for success, 1 for a failure found, 2 for a usage error.
- **File header:** `#!/usr/bin/env python3` on line 1, `# -*- coding: utf-8 -*-` on line 2.
- **Calling a script:** from a skill, as in the base SKILL.md; from a mise task, by its repo path, `python3 .claude/skills/pywf-oss-base/scripts/<script_name>.py`.
- `assets/` holds templates with placeholders. They are not valid Python, so ruff excludes them.
---
## 5. After any change to the family
1. Run `python ${CLAUDE_SKILL_DIR}/scripts/update_skill_index.py` to regenerate the references index in SKILL.md (`--check` only verifies it).
2. Keep three places in agreement:
- each invoker's "Read" list;
- the "Reads from this base" column of the "Which skill to use" table in the skill-family workflow reference;
- the handoff table in the same reference.
For a new or renamed skill, also update the skills table in `CLAUDE.md`.
3. Run `python ${CLAUDE_SKILL_DIR}/scripts/check_skill_family.py --skills_dir .claude/skills`, with `--forbid <forbidden_name>` for each project name that must not appear (such as the seed repo's demo package). It checks that every frontmatter is well formed, every `../pywf-oss-base/...` path resolves, every cited reference exists, no reference cites a later one, and no forbidden name appears.
4. Run `mise run lint`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |