Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
The Skill Map
A living reference for the Agent Skills ecosystem: 4,902 skills across 39 repositories, 13 domains, 5 uncharted territories
An interactive force-directed graph of the Claude / agent skills ecosystem, backed by a crawler that has swept 39 public repositories and collected 4,902 SKILL.md files in full. Thirteen domains. Hundreds of organizations. And five territories with zero published skills: the negative space that tells you where the ecosystem is missing something.
The Session Layer Gap
Every published skill in this corpus does one thing: it makes Claude better at a task. Write better Terraform. Review smarter contracts. Generate cleaner dbt models. The task layer is well-covered: 4,902 skills and counting.
Nobody has published a skill for how to run a day.
That's the session layer: the skills that manage continuity across a working session. Captain's log. Day planning. Context snapshots. Weekly resets. Reading list ingestion. These aren't task-layer utilities: they're the infrastructure that makes Claude a persistent collaborator rather than a stateless tool. Stateless tools get abandoned. Persistent collaborators become habits.
DHK has been building and running session-layer skills in production for over a year, before Anthropic formalized the Skills format in October 2025. None of that work appears in the public corpus, because nobody had a category for it. This map makes the gap visible.
The Negative Space
Five territories with zero published skills as of June 2026:
1. The Session Layer: Skills for continuity: captain's log, day planning, context handoff, weekly resets. The stickiest use case in the ecosystem, and the most underbuilt.
2. Claude as Personal OS: Orchestrated workflow systems that embed Claude into daily rhythm. Not individual automations: a coordinated layer that runs your day.
3. Skill Discovery: No canonical registry. No versioning standard. No dependency management. The npm problem, unsolved.
4. Skill Evaluation: No benchmark. No eval harness. Skills ship on vibes. The ecosystem has no way to measure whether a skill is good.
5. Healthcare & Life Sciences: The largest enterprise vertical in software. Zero published skills. HIPAA compliance, clinical workflow, EHR integration: all dark matter.
How to Use the Map
- Click a domain node to filter to that domain and surface the persona who lives there
- Click an org node to see all skills from that organization
- Click a skill node to see the skill's name, domain, and org
- Click "DHK" in the filter bar to isolate the session-layer cluster in the negative space
- Click "Uncharted" to highlight the five red clouds: the territories with no published skills
- Click any red cloud to open the territory description
Data Sources
The map is built from two layers of data:
Interactive graph (what you see at the link above): 1,121 skills across 53 organizations, sourced from VoltAgent/awesome-agent-skills — a curated index of published agent skills. 229 nodes, 306 links, 13 domains.
Raw corpus (powering future analysis): a GitHub crawler has swept 39 repositories and collected 4,902 Claude-format SKILL.md files in full — plus 418 Gemini-format files catalogued by metadata. That's the actual content of the skills, not just the index entries.
Largest repositories in the corpus
| Repository | Skills | Stars | Focus |
|---|---|---|---|
| BbgnsurfTech/claude-skills-collection | 1,337 | 20 | Comprehensive skills + plugins collection |
| affaan-m/ECC | 881 | 221k | Agent harness + skills, multi-framework |
| davila7/claude-code-templates | 872 | 28k | Claude Code configuration and templates |
| alirezarezvani/claude-skills | 345 | 19k | 337 skills across 30+ agent types |
| borghei/claude-skills | 340 | 300 | 338 skills across 16 domains |
| wshobson/agents | 156 | 37k | Multi-harness plugin marketplace |
- Crawler:
crawlers/crawl.py— repo-scoped git tree walks, SHA-diff incrementality, source quality metrics - Seed lists:
crawlers/crawl-lists/— curated and research-sourced repo lists - Last crawled: June 2026
- Domains: 13 (Developer Tools, Security & Auth, Frontend & Design, Agent Orchestration, Data & Databases, DevOps & Infrastructure, Testing & Debugging, Document Creation, Marketing & Content, Product Management, Finance & Payments, Communication, Media & Creative)
- Negative space: 5 uncharted territories identified by DHK
About
Built by DHK: practitioner writing on data, AI, and working systems.
- Substack: dhkondata.substack.com
- GitHub: github.com/dhk
The argument behind the map: The Session Layer: published June 2026.