Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →tschuehly/embabel-agent-builder
Use only when the user explicitly invokes the embabel-agent-builder skill by name or path. When invoked, route the request to the right Embabel agent-building phase: planning from a prompt/product/feature spec, current-docs research for implementation questions, implementation from an approved high-level spec, or an end-to-end spec-first build.
Passed 9 of 16 test cases with the skill, and 4 of 16 without it.
and 6 more in the full report
8 fixed5 passed either way9 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
We need to lookup version-matched documentation topics for Embabel RAG retrievers. Which map reference should embabel-agent-builder load?
| Checked | Without skill | With skill |
|---|---|---|
| Loads references/docs-map.md | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill only when the user explicitly invokes embabel-agent-builder by name or path. Do not use it automatically for general Embabel work.
This skill routes a prompt, product spec, feature spec, docs question, existing high-level agent spec, or implementation request to the right Embabel workflow. The default route is spec-first: produce or update a high-level Embabel agent spec, then use that spec to ground implementation and progress.
Choose one route from the user's input:
| Input | Route | Load |
|---|---|---|
| Prompt, product spec, feature spec, ticket, rough idea, or existing domain model | Planning | references/planning.md |
| Specific Embabel API, annotation, tool, state, RAG, chatbot, planner, invocation, or testing question | Docs research | references/docs-research.md |
| Approved high-level Embabel agent spec, or request to implement an existing spec | Implementation | references/implementation.md |
| Request to build the whole agent from idea/spec through code | End-to-end | Planning, then Docs research as needed, then Implementation |
| Ambiguous request | Clarify route | Ask whether the user wants planning only, docs research only, implementation from a spec, or end-to-end build |
Do not start implementation without a high-level Embabel agent spec unless the user explicitly says to bypass planning. If the user bypasses planning, state that this is risky because Embabel's typed GOAP flow depends on domain and blackboard design.
Inspect the codebase before asking questions that local files can answer. Look for existing domain types, services, agents, tests, docs, CONTEXT.md, CONTEXT-MAP.md, and ADRs.
Challenge vague or overloaded terms. Propose canonical domain language when needed.
Ask one unresolved blocking question at a time, give a recommended answer, and continue only after the user accepts, rejects, or refines it.
When implementation touches Embabel-specific APIs, fetch version-matched docs using references/docs-map.md and scripts/fetch-docs-section.py. Run the fetcher from the target project or pass --project-dir so it resolves the repository's Embabel version. Fetch only the narrowest docs needed.
Do not inspect Maven JARs, source JARs, decompiled classes, or local dependency caches as a substitute for Embabel docs research. Use them only after the version-matched docs are insufficient and there is a concrete compile/runtime mismatch to investigate.
Do not copy migration, roadmap, or old-vs-new framing into project documentation. Describe only the supported current workflow and intended state.
For every LLM action that creates a typed object, require a structured-output contract. Cover the target output type, nullable fields, required Kotlin constructor parameters, SomeOf fields, absent-value representation, validation/defaulting strategy, typed examples for strict or optional nested output, and downstream formatter inputs. Do not mark implementation complete while structured-output binding or live/fixture acceptance remains unverified unless the spec records a concrete blocker.
Planning and implementation routes revolve around a high-level Embabel agent spec. Keep it as the grounding artifact and progress tracker.
test cases passed, vs 4 / 16 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.