Add CLAUDE.md with development workflow and quality guidelines

[abotescu20-maker]

Summary

• Adds a comprehensive CLAUDE.md to the project root with development workflow documentation, skill quality guidelines, and cross-reference patterns
• Includes planning strategy, SKILL.md template with size guidelines, description writing guide, MCP server reference table, and a lessons-learned memory section

Test plan

• Verify YAML frontmatter examples in CLAUDE.md render correctly
• Confirm cross-skill references match existing skill names in the repo
• Validate MCP server table against actual configured servers

🤖 Generated with Claude Code

[Gonzih]

The CLAUDE.md content is largely good — the development workflow, quality checklist, description writing guide, and lessons-learned section are all useful. However, there are a few issues that should be addressed before merging:

1. Unrelated changes bundled in the PR

The diff includes modifications to .claude-plugin/marketplace.json (version bump from 2.20.0 → 2.24.0, and registration of three new skills: copilot-skills-integration, infographics, neuropixels-analysis, browser-automation, multi-agent-research) and substantial changes to scientific-skills/adaptyv/SKILL.md. None of these are related to adding CLAUDE.md. Please split these out into a separate PR or explain why they belong here.

2. Marketplace.json changes reference skills with no corresponding SKILL.md in this PR

The new entries (copilot-skills-integration, infographics, neuropixels-analysis, browser-automation, multi-agent-research) are registered in marketplace.json but their skill directories and SKILL.md files do not appear in this diff. If those directories don't exist on main, the registry will point to non-existent paths — which is exactly the anti-pattern CLAUDE.md itself warns against.

3. Minor: commit style inconsistency

CLAUDE.md explicitly says "Do NOT use conventional commits (no feat:, fix: prefixes)" — but the PR title itself doesn't follow the documented style either. Not blocking, but worth noting for consistency.

4. The MCP server table should be verified

CLAUDE.md includes a table of MCP servers available in the environment. The PR description's own test plan acknowledges these need to be validated. Please confirm these entries are accurate for contributors working outside K-Dense's internal environment, or add a note that this table reflects K-Dense's internal setup.

Once the unrelated changes are split out and the marketplace.json integrity is verified, this will be a valuable addition.

[Gonzih]

The CLAUDE.md content itself is high quality — the workflow guidance, SKILL.md template, common-mistakes section, and size guidelines are all useful additions. The license corrections across existing skills are also clearly correct. However, there are a few concerns:

1. Omnibus PR — please split
This PR mixes at least four distinct concerns: (a) CLAUDE.md addition, (b) license field fixes across ~20 existing skills, (c) adaptyv SKILL.md improvements, and (d) three brand-new skills with marketplace registration. Each of these would be a clean, reviewable unit on its own. Bundling them makes the diff hard to reason about and means a problem in one area blocks everything else.

2. Marketplace version jump
marketplace.json version goes from 2.20.0 → 2.24.0 — a jump of 4 minor versions for what appears to be 3 new skills. Please align the version bump with the actual number of skills added (one minor version per skill is the typical convention here, so 2.23.0 would be expected for 3 additions).

3. copilot-skills-integration skill seems off-topic
A skill describing cross-compatibility with GitHub Copilot Agent Skills and M365 Copilot feels out of scope for a scientific-skills library. This repo is focused on domain skills for scientific computing — meta-skills about other AI platforms don't belong here without a clear rationale. Please either justify its inclusion or move it to a separate PR with discussion.

4. Minor: CLAUDE.md cross-skill reference map
A few entries in the cross-skill reference map point to skills not yet in the repo (e.g. neuropixels-analysis). Consider only listing skills that actually exist at merge time to avoid broken references.