Skip to content

CLAUDE.md

Latest commit

 

History

History
61 lines (42 loc) · 3.66 KB

CLAUDE.md

File metadata and controls

61 lines (42 loc) · 3.66 KB

Scientific Agentic Engineering

Project

2026 BSSw Fellowship project by Victor Weeks (NSF NCAR). Developing methodology for integrating AI agents into scientific software workflows.

Architecture: Bare Astro + Tailwind CSS + Pagefind + Content Collections. No Starlight governance. Quarto used only for interactive notebook tutorials (M2+). Each milestone extends the component library — the site grows incrementally from a simple foundation.

Plan: See docs/plan.md for the current implementation plan with milestone roadmap and file details.

Sensitive Content — DO NOT COMMIT

The SOW document lives one directory up (../2026_BSSw-SOW_VictorWeeks_revised_20260212.docx). Never reference specific financial amounts, SOW text, or budget details in any committed file. The .gitignore blocks *.docx, SOW*, budget*, etc., but remain vigilant.

Development Log (docs/development-log.md)

This file transparently records how AI-assisted development was used to build this project. It is a pedagogical artifact — the site is about responsible AI-assisted development, and the log demonstrates that methodology in practice.

When to update the development log

Update the log after every significant interaction — defined as any interaction that:

  • Makes an architectural or design decision
  • Creates, modifies, or deletes files in the repo
  • Involves research that informed a decision
  • Surfaces a pattern worth documenting (AI behavior, human override, quality issue)
  • Results in a rejected approach (failed attempts are as educational as successes)

Do NOT log routine, mechanical actions (typo fixes, formatting, etc.) unless they reveal something interesting.

How to update the development log

  1. Append a new entry at the end of the current session section, before the "Template for Future Entries" section
  2. Use the template from the bottom of the log file:
    • ### Interaction N: [Title] — increment N from the last entry
    • **Human prompt (paraphrased):** — what was asked (paraphrase, do not copy verbatim)
    • **What the AI did:** — tools used, agents launched, files created/modified
    • **Key findings/output:** — what was produced or discovered
    • **Human decisions:** — what the human chose, modified, or rejected
    • **Observation:** — what this reveals about AI-assisted development
  3. Start a new session section (## Session N — YYYY-MM-DD: [Theme]) when the conversation is a new Claude Code session
  4. Be honest about AI missteps, over-engineering, and human corrections — these are the most pedagogically valuable entries
  5. Update the "Patterns Observed" section when a new pattern emerges across multiple interactions

Tone

Write as a neutral technical observer, not as marketing. The log should read like a well-kept lab notebook: factual, specific, reflective. Attribute decisions clearly — who decided what and why.

Milestones

  • M1 (due July 2026): Landing page + Framework document layout + Framework draft
  • M2 (due Oct 2026): Catalog + Tutorial layouts, Quarto/Pyodide integration, US-RSE'26
  • M3 (due Jan 2027): Failure mode exhibits + Full tutorial series
  • M4 (due April 2027): Workshop recordings, HPC webinar, BSSw.io blog post

Conventions

  • Licensing: Content is CC-BY-4.0 (LICENSE-CONTENT), code is MIT (LICENSE-CODE)
  • BSSw acknowledgment: Must appear on every page footer and in README
  • Component naming: PascalCase .astro files in src/components/
  • Content Collections: Typed schemas in src/content/config.ts, one collection per deliverable type
  • Commit style: Focused commits; split foundation/structure/CI into separate commits
⚡