| name | docs-authoring-review |
| description | Comprehensive guidelines for writing, auditing, and reviewing documentation in torchfits. Enforces human-first writing, zero AI/developer jargon, technical precision against src/, clean Mermaid rendering, and multi-tier verification. |
| disable-model-invocation | true |
Documentation Authoring & Review Guidelines (torchfits)
Use when authoring new documentation pages, updating existing guides, or conducting documentation audits.
1. Core Principles
- Human-First & Direct: Write for astronomers, astrophysicists, and machine learning researchers. Lead with direct, runnable Python and CLI examples rather than speculative boilerplate.
- Zero AI / Developer Jargon: Strip out internal development terms from user-facing pages:
- ❌ "smoke tests", "tax", "lanes", "inventory", "coding agents"
- ✅ "validation", "startup overhead", "environments/targets", "inspection", "developers"
- Faithful to Implementation: Every function signature, parameter name, CLI flag, and default value must strictly match the code in
src/torchfits/.
- Astronomical Context: Use realistic astronomical examples (e.g. multi-band survey cutouts, SDSS/DESI spectra, Chandra event tables, CFHT mosaics) rather than toy random arrays where helpful.
2. Formatting & Syntax Rules
- Clickable Links: Use relative Markdown links with explicit anchor IDs where needed.
- Mermaid Diagrams:
- Quote all node labels containing special characters, brackets, or equations:
node_id["Label Text (Extra Info)"].
- Prefer clean, structured flowcharts (
flowchart LR or flowchart TD) and sequence diagrams.
- Code Blocks: Use fenced code blocks with appropriate language tags (
python, bash, text, yaml, mermaid). All Python snippets in docstrings and Markdown are formatted and verified via Ruff.
3. Verification & Gating
Before submitting any documentation changes, execute all verification tiers:
pixi run docs-contract
pixi run docs-links
pixi run preflight-push