What Changed on ASE This Week: Industry Mappings, Source-Aligned Stars, and New Operator Skills
In Short
- ASE’s industry collections got tighter this week, with clearer boundaries between AI Agency, GTM and RevOps, Product Analytics, Education, and other operator-facing groups.
- Several new skills landed in practical places where teams can actually use them: evaluation, RAG, supervised agent work, MCP interfaces, and local model testing.
- The star and metadata work stayed source-aligned. When ASE cannot prove a repo, package, or download count from the published source, it leaves the field blank instead of guessing.
- For operators, the practical takeaway is simple: browse by workflow first, then use trust signals and source-backed metadata to decide what deserves a pilot.
| Change | Why it matters | What to try |
|---|---|---|
| Sharper industry mappings | Collections are less likely to mix generic automation with domain-specific work. | Start from Industry Collections before searching broadly. |
| New operator skills | Recent additions cover evaluation, RAG, supervised sessions, and local model endpoints. | Pick one workflow gap and test one skill with a small reviewable task. |
| Source-aligned metadata | Stars, packages, and provenance are useful only when they map to the right upstream project. | Use blank fields as a review cue, not as a reason to invent numbers. |
This week’s ASE work was mostly about making the catalog more useful for people choosing agent workflows under real constraints. The visible result is not one giant feature. It is a set of small improvements that reduce the chance of picking a skill because it appeared in the wrong place, had an unverified signal, or looked more mature than its source actually supports.
That matters because agent skill selection is becoming an operations problem. A team does not only ask, “Can this agent do the task?” It asks whether the skill fits the workflow, whether the source looks inspectable, whether the metadata is backed by the right project, and whether there is a human review path before anything sensitive changes.
The weekly pattern also makes ASE easier to audit. When collection changes, source updates, and new skills are described in operator terms, readers can see what changed without reverse-engineering internal maintenance work.
Who this is for
This update is for builders and operators using ASE to shortlist skills for internal pilots. It is especially relevant if you browse by industry, compare skills by source signals, or maintain a small agent stack for product analytics, research, client implementation, or support work.
If you are new to the catalog, start with the skills browse page or the Industry Collections page. If you already follow ASE, the useful part of this week’s work is the direction of travel: fewer noisy placements, more practical routing, and continued refusal to fill metadata gaps with guesses. That same principle showed up in the recent post on why blank is better than wrong for agent skill metadata.
Starter workflow
Use this week’s changes as a short selection workflow.
- Choose the business surface first. If the work belongs to product analytics, education and research, AI agency delivery, finance, legal, support, or another vertical, begin with the relevant industry collection instead of a keyword search.
- Open two or three candidate skills and check whether the title, source, and described workflow match the job you need done. A good match should mention the system or work object you actually use.
- Look at trust signals after fit, not before it. GitHub stars and package metadata help only when they are tied to the correct upstream source.
- Run a small pilot with reviewable output: a trace, summary, evaluation report, RAG answer with source context, or local endpoint test. Avoid starting with broad autonomous permissions.
- Document the result. Keep the skill if it saved review time or improved evidence quality. Drop it if it only produced a nicer-looking version of manual work.
Recommended ASE skills
| Skill | Good first use | Operator note |
|---|---|---|
| Monitor and evaluate LLM agent traffic with Helicone | Review agent calls, costs, latency, and evaluation traces. | Useful for product teams that need evidence before expanding an agent workflow. |
| Use Prompt Flow for LLM workflow testing and evaluation | Test prompt and chain behavior before shipping changes. | Good fit when repeatable evaluation matters more than one-off demos. |
| Build graph RAG context with Neo4j LLM Graph Builder | Turn documents into graph-backed retrieval context. | Best for knowledge-heavy teams that need relationships, not just chunks. |
| Run RAG chatbot workflows with Verba | Prototype a reviewable RAG chatbot over a bounded corpus. | Start with internal documentation or research notes before user-facing use. |
| Run supervised Suna sessions for reviewable agent work | Keep autonomous work observable while an operator supervises progress. | Good for AI agency and field-engineering style delivery where review matters. |
| Render interactive MCP tool UIs with mcp-ui | Expose richer tool interactions instead of plain text-only controls. | Useful when the operator needs to inspect or guide tool state during a task. |
| Serve local model endpoints for agent tests with OpenLLM | Run local model endpoints for repeatable agent tests. | Useful when teams want isolated tests before introducing hosted dependencies. |
What changed in the collections
The industry mapping work this week focused on practical fit. AI Agency and field-engineering style work now leans more toward client implementation, visual workflow orchestration, supervised sessions, and reviewable delivery. GTM and RevOps keeps the marketing, lifecycle, CRM, scheduling, and revenue-operations workflows where they belong. Product Analytics and Growth Ops gained stronger evaluation and observability paths through skills like Helicone and Prompt Flow.
Education and research also picked up stronger RAG coverage through Neo4j LLM Graph Builder and Verba. Those placements matter because they help readers distinguish research-context assembly from general chatbot tooling. The catalog is more useful when a collection says, “try this for this kind of work,” not merely, “this tool mentions AI.”
What to watch
- Do not treat a high star count as a workflow recommendation. Stars are an adoption signal, not a fit signal.
- Watch for collection boundaries. If a skill appears in an industry page, it should map to the domain workflow, not just to generic automation.
- Blank metadata is intentional when ASE cannot prove the source. That is a quality control choice.
- Recent skills are good pilot candidates, but they still need local review against your permissions, data boundaries, and failure modes.
FAQ
Did ASE add a new industry collection this week?
The bigger change was refinement of existing and recently expanded collections. The goal was cleaner placement, especially around AI Agency, GTM and RevOps, Product Analytics, and Education and Research workflows.
Should I choose skills by GitHub stars?
No. Use stars as one signal after you confirm workflow fit, source alignment, maintenance posture, and review requirements. A popular upstream project can still be the wrong skill for your task.
Why does ASE leave some metadata blank?
Because wrong provenance is worse than missing provenance. If a package, repo, or download count cannot be tied confidently to the published skill, ASE should leave the field empty until the source supports it.
What is the fastest way to use this week’s update?
Pick one workflow you already run manually, open the closest industry collection, choose one recent skill, and test it on a narrow task with human review before widening permissions.