docs: add skill cards for the five complexa-* agent skills

[trvachov]

Adds a skill-card.md alongside each of the five complexa-* skills, following the NVIDIA skill card format used across the NVIDIA/skills catalog (326/326 skills there carry one).

Why

The NVIDIA/skills sync workflow drops any skill missing skill-card.md, skill.oms.sig, or an eval dataset — the enforcement step removes the skill dir before the PR is created, so non-compliant skills never merge into the catalog. The same requirement is tracked as SRC-11 in bionemo-agent-toolkit's CONTRIBUTING.

State of this repo (dev) against those three artifacts:

Artifact	Status
eval dataset	not on dev — arrives with #57 (aggregator-compat)
skill-card.md	this PR
skill.oms.sig	outstanding — needs nvskills-ci wired here

The cards state the eval gap explicitly rather than claiming coverage that isn't on this branch. Once #57 merges, those lines should be updated to point at the landed evals/evals.json.

What's in the cards

Per skill: description, owner, license, use case, requirements (complexa CLI, .env, GPU class and VRAM, folding-backend weights), risks and mitigations, references, output types, evaluation, and ethical considerations.

Risk sections are specific to each skill — in-silico metrics not predicting experimental binding (complexa-design), evaluating designs with the same model family that generated them (complexa-evaluate-pdbs), multiplicative GPU cost from cartesian-product expansion (complexa-sweep), hotspot residue-numbering errors silently redirecting a whole campaign (complexa-target), and third-party checkpoint licensing (complexa-setup).

complexa-design and complexa-sweep carry explicit biosecurity language, since those skills generate novel protein sequences at scale: users are responsible for screening designed sequences before synthesis and for compliance with biosafety review and export-control obligations.

Evaluation results are marked pending, not fabricated. NVSkills-Eval has not been run against these skills.

Second commit: a .gitignore fix worth looking at

.gitignore:95 ignores .claude/ wholesale, but the skills under .claude/skills/ are tracked and are vendored downstream. Because a bare .claude/ excludes the directory, git never descends into it and no ! negation can re-include anything below it — so every new file added to a skill dir is silently ignored unless the author remembers git add -f. I hit this writing these cards.

That means an evals/ dir or a reference file added to a Complexa skill can be quietly lost. The fix switches to .claude/* plus !.claude/skills/. Verified both directions: a new file under .claude/skills/ is no longer ignored, and other .claude/ content still is.

It's a separate commit — drop it if the wholesale ignore is deliberate.

Context

bionemo-agent-toolkit vendors these five skills via components.d/proteina-complexa.yml and rsyncs with --delete, so a card committed there is wiped on the next nightly sync. This repo is the source of truth. The catalog carries an interim copy of these same cards in a compliance.d/ overlay, which is retired automatically once this merges.

Targeting dev rather than aggregator-compat since that branch is deleted when #57 merges.

Please correct anything I got wrong about the pipelines or their risks — I wrote these from the SKILL.md files and repo layout, not from running the design pipeline.

🤖 Generated with Claude Code