readme-enrichment - SKILL.md Agent Skill

name: readme-enrichment version: 1.0.0 description: > Enrich README.md with verified hyperlinks for all libraries, papers, models, and standards. Zero-tolerance hallucination policy — every URL is web-verified before insertion. Use when: "update README links", "enrich README", "add hyperlinks to README", "link papers in README", "fix missing links". Do NOT use for: code implementation (use self-learning-iterative-coder), literature reports (use create-literature-report), knowledge graph sync (use kg-sync). last_updated: 2026-03-25 activation: manual invocation: /readme-enrichment metadata: category: documentation tags: [readme, hyperlinks, citations, verification, documentation] relations: compose_with: [create-literature-report] depend_on: [] similar_to: [] belong_to: []

README Enrichment Skill

Systematic enrichment of README.md with verified hyperlinks. Covers 4 categories:

Libraries/Tools — homepage URLs (e.g., [Prefect](https://prefect.io/))
Papers/Models — DOI or ArXiv links (e.g., [SAM3](https://arxiv.org/abs/...))
Standards/Regulations — official pages (e.g., [IEC 62304](https://...))
Issues/Planned — GitHub issue cross-references (e.g., [#894](link))

Core Rules

R1: Zero Hallucination Policy (NON-NEGOTIABLE)

Every URL inserted into the README MUST be verified via WebFetch or WebSearch before insertion. "I'm fairly confident this is the URL" is NOT acceptable. The cost of a broken link in a public README is higher than the cost of verification.

R2: Bibliography First

Before web-searching for a paper URL, check these local sources:

knowledge-graph/bibliography.yaml — canonical citation database
/home/petteri/Dropbox/github-personal/sci-llm-writer/biblio/biblio-vascular/ — literature files

R3: Link Format Convention

Libraries: [ToolName](homepage) — name only, no version
Papers: [ModelName](DOI_or_ArXiv) — model/method name as anchor, NOT "Author et al." This keeps the README clean: [SAM3](https://arxiv.org/abs/2408.12166) not [Ravi et al. (2025)](https://arxiv.org/abs/2408.12166)
Standards: [IEC 62304](URL) — standard number as anchor
Issues: [#NNN](https://github.com/petteriTeikari/vascadia/issues/NNN)

R4: Preserve Existing Correct Links

Never modify a link that is already correct. Only ADD missing links or FIX broken ones.

R5: Science Issues as Grouped Link

In the "Planned" section, group science-tagged open issues as a bullet with individual links: Science backlog: [#894](url), [#895](url), ...

4-Phase Pipeline

Phase 1: EXTRACT (read-only)

Parse README.md and identify every unlinked reference:

Tools/libraries mentioned without [Name](URL) syntax
Papers/models cited without DOI/ArXiv
Standards without official page links
Planned items without issue cross-refs

Phase 2: RESOLVE (research)

For each unlinked reference:

Check bibliography.yaml first
Check sci-llm-writer/biblio/biblio-vascular/ second
Web search ONLY if not found locally
Record: {name, url, source, verified}

Phase 3: VERIFY (zero-tolerance)

For every resolved URL:

WebFetch the URL and confirm it returns 200 + correct content
For ArXiv: verify the paper title matches
For GitHub: verify the repo exists
For DOI: verify it resolves

Phase 4: APPLY (edit README)

Apply all verified links to README.md:

Use Edit tool for surgical replacements
Preserve existing correct links (R4)
Add science issue group to Planned section
Run ruff format check after edits

Anti-Patterns (BANNED)

Memory-based URLs: "I remember PyTorch is at pytorch.org" → VERIFY FIRST
Guessed DOIs: constructing DOIs from patterns → ALWAYS web-verify
Batch insertion without verification: inserting 20 links then checking → verify EACH
Removing existing content: enrichment is ADDITIVE, never destructive