Skip to content

Add programmer skill: senior-engineer judgment from 8 SWE books#184

Open
Chenwei-1999 wants to merge 2 commits into
K-Dense-AI:mainfrom
Chenwei-1999:add-programmer-skill
Open

Add programmer skill: senior-engineer judgment from 8 SWE books#184
Chenwei-1999 wants to merge 2 commits into
K-Dense-AI:mainfrom
Chenwei-1999:add-programmer-skill

Conversation

@Chenwei-1999

Copy link
Copy Markdown

What

Adds a new skill at skills/programmer/ — a portable, agent-neutral skill that gives a coding agent senior-engineer judgment across naming, functions, comments, error handling, complexity management, module/OO design, SOLID, architecture boundaries, data systems, concurrency, testing, code review, long-term engineering trade-offs, and idiomatic Python.

Why it fits this repo

Scientific software is read and changed far more than it is written — reproducible pipelines, data-intensive workflows, and shared lab codebases live for years. This skill is the cross-cutting code-quality layer for that work: it helps an agent write and review the engineering around scientific packages (e.g. structuring an anndata/scanpy pipeline, designing storage/replication trade-offs à la DDIA), complementing the domain skills already here.

Design

  • Source-tagged rules. Every rule is tagged with the canonical book(s) behind it (A Philosophy of Software Design, Clean Code, Code Complete 2e, Clean Architecture, Designing Data-Intensive Applications, Software Engineering at Google, The Pragmatic Programmer, Fluent Python 2e). Where books converge, that's the signal to trust; where they conflict (e.g. many-small-classes vs. deep modules), the skill takes a reconciled stance.
  • Two-tier. Tier 1 is always-available embedded judgment (SKILL.md + per-book catalogs in references/). Tier 2 is a deliberate path to primary sources for high-stakes or contested calls.
  • No copyrighted PDFs shipped. The references/ catalogs are original distillations; source texts stay with their owners. A nested .gitignore keeps any on-demand-fetched book PDF out of the repo.

Format / spec conformance

  • name: programmer matches the directory name.
  • metadata.version: "1.0" present and quoted (new skill).
  • SKILL.md is 324 lines (under the 500-line recommendation); long material lives one level deep in references/.
  • No secrets, credentials, or private data.

Layout

skills/programmer/
  SKILL.md                 # router + judgment, rules tagged by book
  references/              # 11 per-book / per-runtime catalogs
  scripts/read_book.py     # list/read/search a supplied book PDF (Tier 2)
  assets/README.md         # how book PDFs are supplied (never committed)
  .gitignore               # keeps fetched PDFs / pycache out of the repo

Testing

Frontmatter conforms to the required Agent Skills format and this repo's metadata.version convention. scripts/read_book.py is a self-contained CLI (--list / --query) over an extractor with pdftotext/pypdf/PyPDF2/built-in fallbacks. The skills-ref / skill-scanner validators were not run in the authoring environment; happy to address any validator findings in review.

A portable, agent-neutral skill giving a coding agent senior-engineer
judgment across naming, functions, comments, error handling, complexity,
module/OO design, SOLID, architecture boundaries, data systems, concurrency,
testing, code review, and idiomatic Python. Each rule is tagged with the
book(s) behind it.

Two-tier design: Tier 1 is embedded distilled judgment (SKILL.md plus
per-book reference catalogs); Tier 2 fetches the primary source on demand
for high-stakes or contested calls. No book PDFs are shipped -- the catalogs
are original distillations and the source texts remain with their owners.

Conforms to the Agent Skills frontmatter spec: name matches the directory
and metadata.version is "1.0".

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b10397731e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread skills/programmer/SKILL.md Outdated
Address Codex review on PR K-Dense-AI#184. The router-level Tier-2 instruction told the
agent to fetch a book PDF without the authorization constraint that
assets/README.md and the Additional Resources section already carry, so on a
network-enabled runner the agent could auto-download copyrighted books --
undermining the skill's stated no-copyrighted-PDFs design. Align the primary
instruction to require a copy the reader is entitled to from a legitimate
source, matching the other two fetch mentions.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant