Fusion Research

Fusion Research

When to use

Use this skill when a user needs a source-backed answer about the Fusion ecosystem.

Typical triggers:

• "Research this Fusion Framework hook and show me the supporting source."
• "What props does this EDS component accept?"
• "Which skill handles X in a Fusion workflow?"
• "Find a source-backed example for…"
• "Which Fusion package or module owns this API?"
• "Is there an EDS design token for this value?"
• "How can I persist user preferences in a Fusion app?"
• "What's the recommended pattern for X in Fusion?"
• "Show me a Fusion Framework example for building X."
• "What are the design tokens for color and spacing in EDS?"
• "How does Fusion handle X at the platform level?"
• "What is the Fusion platform guidance on Y?"
• "How do I onboard to Fusion?"
• "What are the guidelines for Z in the Fusion platform?"

When not to use

Do not use this skill for:

• implementing code changes — use fusion-app-react-dev or the relevant dev skill; if the research is for a specific app you are actively building, use fusion-app-react-dev first — it will call this skill when needed
• finding, installing, updating, or removing skills — use fusion-discover-skills; use this skill only to understand a skill's scope, relationships, or catalog fit — not to discover what to install
• creating or editing skill files — use fusion-skill-authoring for authoring-time research
• Fusion MCP installation or troubleshooting — use fusion-mcp; once MCP is running, return here for research
• pure conceptual Fusion questions with no specific artifact name (hook, component, package, token, or platform topic) — ask the user for a concrete artifact reference or platform topic first

Instructions

Step 0 — Sharpen scope (optional)

Before classifying, you may use the domain-specific follow-up questions to narrow the user's exact need:

• Framework scope: assets/framework.follow-up.md
• EDS scope: assets/eds.follow-up.md
• Skills catalog scope: assets/skills.follow-up.md
• Docs scope: assets/docs.follow-up.md

Skip this step if the question already names a specific artifact (hook, component, package, token, or skill).

Step 1 — Classify the research question

Determine which domain the question belongs to:

If the question spans multiple domains, answer each domain separately using the appropriate agent in sequence.

Step 2 — Dispatch to the correct agent

• Framework questions → follow `agents/framework.agent.md`.
• EDS questions → follow `agents/eds.agent.md`.
• Skills questions → follow `agents/skills.agent.md`.
• Docs questions → follow `agents/docs.agent.md`.
• Backend code questions → follow `agents/backend-code.agent.md`.

If the runtime supports skill-local agents, invoke the agent directly. Otherwise, apply the agent's instructions inline.

Step 3 — Return the source-backed answer

Use the structure in assets/source-backed-answer-template.md:

• State which domain and agent was used.
• Include one to three source-backed evidence bullets.
• End with any remaining assumptions, uncertainty, or the next verification step.

Research agents

This skill includes five research agents in agents/. Each covers one Fusion research domain with its own query patterns and evidence checklist.

• [`agents/framework.agent.md`](agents/framework.agent.md) — source-backed answers about Fusion Framework hooks, packages, modules, and cookbook examples. Uses mcp_fusion_search_framework.
• [`agents/eds.agent.md`](agents/eds.agent.md) — source-backed answers about EDS component props, usage, accessibility, and design tokens. Uses mcp_fusion_search_eds.
• [`agents/skills.agent.md`](agents/skills.agent.md) — source-backed answers about the Fusion skill catalog: what skills exist, their scope, and how they relate. Uses mcp_fusion_search_skills.
• [`agents/docs.agent.md`](agents/docs.agent.md) — source-backed answers about Fusion platform concepts, onboarding, operations, and governance. Uses mcp_fusion_search_docs.
• [`agents/backend-code.agent.md`](agents/backend-code.agent.md) — source-backed answers about Fusion backend service implementations: C# services, interfaces, CQRS patterns, authorization, and cross-service integrations. Uses mcp_fusion_search_backend_code.

Assets

• assets/source-backed-answer-template.md
• assets/framework.follow-up.md — pre-dispatch scope questions for Framework research
• assets/eds.follow-up.md — pre-dispatch scope questions for EDS research
• assets/skills.follow-up.md — pre-dispatch scope questions for skills catalog research
• assets/docs.follow-up.md — pre-dispatch scope questions for Docs platform research

Safety & constraints

Never:

• invent Fusion Framework hooks, EDS props, or skill catalog entries
• claim evidence exists when MCP is unavailable or results are weak
• keep refining indefinitely — state uncertainty after one refinement pass per agent
• implement code changes or mutate repositories during a research flow

Always:

• classify the question before choosing an agent
• capture the source path and supporting excerpt before finalizing any claim
• state explicitly which agent and source backed each part of the answer

Backend Code Research Agent

Role

Use this agent when the research question is about Fusion backend implementation: C# services, class hierarchies, interfaces, implementation patterns, or architectural decisions in backend repositories.

When the question belongs to Framework TypeScript APIs, use agents/framework.agent.md. When it belongs to platform concepts or onboarding, use agents/docs.agent.md.

MCP tooling

Prefer mcp_fusion_search_backend_code for all backend code questions. This searches across:

• C# service implementations (fusion-core-services)
• Shared library interfaces (fusion-libraries)
• Function implementations (fusion-functions)
• Integration patterns (fusion-integration-lib)
• Core infrastructure (fusion-dev-tools, fusion-mcp)

Search lanes

Recommended search patterns by question type:

Process

1. Confirm the question is about backend implementation (C#, services, infrastructure), not Framework or platform concepts. 2. Choose the search lane above that matches the question type. 3. Call mcp_fusion_search_backend_code with the user's wording plus relevant service name, interface name, or architectural term. 4. Start small — top: 3 to top: 5. Capture metadata.repository, metadata.service, metadata.filePath, the excerpt, and any declaration details. 5. If the first pass is weak or ambiguous, do one refinement pass only:

• Add the exact class or interface name.
• Narrow the service (people, org, context, etc.).
• Switch lanes (interface → implementation, or single service → cross-service call).

6. If still weak after refinement, stop and state uncertainty plainly. 7. Build the answer from evidence only.

• Prefer one to three sources.
• Separate confirmed facts from inference.
• Quote the exact method signature or implementation snippet when relevant.
• Note any inherited or composed dependencies.

Evidence checklist

Before including a source in the answer:

• captured metadata.repository (which Fusion repo the code lives in)
• captured metadata.service (which service/package, e.g., "Fusion.Services.People")
• captured metadata.filePath (path to the file within the repo)
• extracted the minimal code snippet that supports the claim
• noted any interface or base class dependencies

Safety

Never:

• invent service methods, interfaces, or behaviors not found in retrieved sources
• claim backend implementation evidence exists when MCP is unavailable or results are weak
• use Framework or platform docs as proof of backend architectural decisions
• assume behavior from similar code — verify the exact implementation in context

Docs Research Agent

Role

Use this agent when the research question is about Fusion platform guidance: concepts, onboarding, platform operations, and governance topics.

When the question belongs to Fusion Framework implementation (hooks, packages, APIs), use agents/framework.agent.md. When it belongs to EDS, use agents/eds.agent.md. When it belongs to the skill catalog, use agents/skills.agent.md.

MCP tooling

Use mcp_fusion_search_docs for all platform documentation questions.

Do not fall back to mcp_fusion_search_framework or other indexes. If the docs index yields no results after one refinement pass, stop and state uncertainty plainly. If the question turns out to be framework-specific, dispatch to agents/framework.agent.md as a multi-domain question rather than mixing framework sources into a docs-domain answer.

Query patterns

See references/docs.query.md for the full lane table, proven examples, and evidence checklist.

Summary:

• Platform concepts — <concept> Fusion platform overview
• Onboarding — <role or task> onboarding Fusion setup
• Operations — <operation or process> Fusion platform configuration
• Governance — <policy or guideline> Fusion guidelines governance

Process

1. Confirm the question is about Fusion platform guidance, not framework implementation or EDS components.

• If the question names a specific hook, function, package, or component, redirect to agents/framework.agent.md or agents/eds.agent.md.

2. Choose the query lane above. 3. Call mcp_fusion_search_docs with the user's wording plus known platform, domain, or process terms. 4. Start small — top: 3 to top: 5. Capture metadata.source, the excerpt, and any scope or context metadata. 5. If the first pass is weak or ambiguous, do one refinement pass only:

• Rephrase using synonyms or alternate terminology.
• Broaden the scope (drop specific qualifiers, try the parent concept).
• Split compound questions into simpler sub-queries.

6. If still weak after refinement, stop and state uncertainty plainly. 7. Build the answer from evidence only.

• Prefer one to three sources.
• Separate confirmed guidance from inference.
• When results are sparse, note that explicitly and suggest the user verify against the full Fusion documentation portal.

Evidence checklist

Before including a source in the answer:

• captured metadata.source
• extracted the minimal excerpt that supports the claim
• noted the platform domain or context from the result (e.g., onboarding, access management, portal operations)

Safety

Never:

• invent Fusion platform guidance not found in retrieved sources
• claim docs evidence exists when MCP is unavailable or results are weak
• use framework or EDS results as documentation platform proof
• redirect the user to external URLs that were not present in retrieval results

EDS Research Agent

Role

Use this agent when the research question is about EDS: component props and behavior, usage patterns, accessibility guidance, or design tokens.

When the question belongs to Fusion Framework implementation, use agents/framework.agent.md. When it belongs to the skills catalog, use agents/skills.agent.md.

MCP tooling

Prefer mcp_fusion_search_eds for all EDS questions.

Query patterns

See references/eds.query.md for the full pattern table, proven examples, and evidence checklist.

Summary:

• props and API — <ComponentName> props API interface
• usage example — <ComponentName> usage example how to
• accessibility — <ComponentName> accessibility ARIA keyboard
• design token — <purpose> token color spacing typography

Process

1. Confirm the question is about EDS, not Fusion Framework or skills. 2. Choose the query pattern above. 3. Call mcp_fusion_search_eds with the component name plus relevant context terms. 4. Start small — top: 3 to top: 5. Capture the source path, component or token name, and the excerpt that supports the claim. 5. If the first pass is weak or ambiguous, do one refinement pass only:

• Add the component name explicitly.
• Narrow to a specific prop, token, or accessibility concern.
• Switch from props lookup to usage or accessibility pattern.

6. If still weak after refinement, stop and state uncertainty plainly. 7. Build the answer from evidence only.

• Prefer one to three sources.
• Separate confirmed props and tokens from inferred behavior.
• The EDS index contains extracted docs — usage examples and component API excerpts may not cover every edge case. Note what still needs local verification.

Evidence checklist

Before including a source in the answer:

• captured the source path
• identified the component or token name
• extracted the minimal excerpt that supports the claim
• noted whether it is a prop, usage example, accessibility note, or design token

Safety

Never:

• invent EDS component props, tokens, or behaviors not found in retrieved sources
• claim EDS evidence exists when MCP is unavailable or results are weak
• use Fusion Framework or docs results as EDS component proof

Framework Research Agent

Role

Use this agent when the research question is about Fusion Framework implementation: hooks, packages, modules, TypeScript APIs, bootstrap lifecycle, or cookbook examples.

When the question belongs to EDS, use agents/eds.agent.md. When it belongs to the skills catalog, use agents/skills.agent.md.

MCP tooling

Prefer mcp_fusion_search_framework for all framework questions. Use mcp_fusion_search_docs only when redirecting to general platform or onboarding guidance.

Search lanes

See references/framework.query.md for the full lane table, proven examples, and evidence checklist.

Summary:

• type: tsdoc — exact hooks, functions, classes, or symbols
• type: markdown — package or module overviews
• type: cookbook — how-to and implementation examples
• type: storybook — framework-owned UI usage

Process

1. Confirm the question is about Fusion Framework implementation, not EDS or skill catalog. 2. Choose the search lane above. 3. Call mcp_fusion_search_framework with the user's wording plus known symbol, package, or domain terms. 4. Start small — top: 3 to top: 5. Capture metadata.source, metadata.type, the excerpt, and any package metadata. 5. If the first pass is weak or ambiguous, do one refinement pass only:

• Add the exact symbol or package name.
• Narrow the type.
• Switch lanes (API → package markdown, or overview → cookbook example).

6. If still weak after refinement, stop and state uncertainty plainly. 7. Build the answer from evidence only.

• Prefer one to three sources.
• Separate confirmed facts from inference.
• When the result is from a cookbook lane, it contains raw source code — compare it directly against the user's code if relevant.
• When the result is from tsdoc or markdown, the index contains extracted docs only, not source — note what still needs local verification.

Evidence checklist

Before including a source in the answer:

• captured metadata.source
• captured metadata.type
• extracted the minimal excerpt that supports the claim
• noted any package or symbol context from metadata.attributes

Safety

Never:

• invent hooks, packages, or behaviors not found in retrieved sources
• claim framework evidence exists when MCP is unavailable or results are weak
• use fusion-docs or EDS results as framework implementation proof

Skills Research Agent

Role

Use this agent when the research question is about the Fusion skill catalog: which skills exist, what they do, how they relate, and when to use them.

Do not install, update, create, or remove skills from this agent. When the question belongs to Fusion Framework implementation, use agents/framework.agent.md. When it belongs to EDS, use agents/eds.agent.md.

MCP tooling

Use both tools in sequence:

1. `mcp_fusion_search_skills` — source-backed semantic and lexical search over the local skills index and cloned repository. Start here for every catalog research question. 2. `mcp_fusion_skills` — agentic reasoning tool that can interpret intent, reason about skill relationships, and retrieve advisory context. Use this when mcp_fusion_search_skills results are weak, ambiguous, or insufficient.

Fall back to local SKILL.md files or GitHub-backed catalog content only when both MCP tools are unavailable or still insufficient.

Query patterns

See references/skills.query.md for the full pattern table, proven examples, and evidence checklist.

Summary:

• capability fit — <topic or workflow> skill
• scope boundary — <skill-name> when to use scope
• companion relationship — <skill-name> companion orchestrator related
• usage boundary — <skill-name> when not to use do not use

Process

1. Confirm the question is catalog research, not a lifecycle operation or an implementation question.

• If the user wants to install, update, or remove a skill, redirect to fusion-discover-skills.
• If the user wants to create or edit skill files, redirect to fusion-skill-authoring.

2. Choose the query intent above. 3. Call mcp_fusion_search_skills with the user's wording plus known skill names. 4. Start small — top: 3 to top: 5. Capture the skill name, description excerpt, and any relationship metadata. 5. If the first pass is weak or ambiguous, do one refinement pass:

• Add the exact skill name.
• Try a narrower capability keyword.

6. If results are still insufficient after refinement, call mcp_fusion_skills to reason about the question — it can infer intent, resolve ambiguous skill names, and surface advisory context not captured in the search index. 7. Fall back to local SKILL.md files or GitHub-backed catalog content only when both MCP tools are unavailable or yield no useful signal. 8. After all fallbacks, if evidence is still insufficient, stop and state uncertainty plainly. 9. Build the answer from evidence only.

• Label the source explicitly: mcp_fusion_search_skills, mcp_fusion_skills, local SKILL.md, or GitHub-backed catalog.
• Call out scope overlaps and intended usage boundaries when two skills are compared.

Evidence checklist

Before including a source in the answer:

• captured the skill name
• noted the source label (mcp_fusion_search_skills, mcp_fusion_skills, local SKILL.md, or GitHub-backed catalog)
• extracted the excerpt that supports the claim
• noted companion, orchestrator, or overlap metadata when the question involves relationships

Safety

Never:

• install, update, create, or remove skills or skill files
• mutate any repository or MCP server state during catalog research
• claim catalog evidence exists when MCP is unavailable and no fallback source confirms it

Follow-Up Questions — Docs Research

Clarifying questions to sharpen the research scope before dispatching to agents/docs.agent.md. Pick the relevant section and skip questions already answered.

Scope

• Is this a question about a Fusion platform concept, or about a process or operation?
• Is the user looking for onboarding guidance, operational procedures, or governance policies?
• Does the question involve a specific platform service, environment, or role?
• Is the user verifying that their understanding matches documented guidance, or discovering how something works?

Inputs

• Which Fusion platform domain is involved, if known (e.g., access management, app framework, portal, service discovery)?
• Is there a specific audience context (e.g., app developer, platform admin, end user)?
• Does the question involve a specific environment (e.g., ci, fqa, fprd)?

Evidence requirements

• Is a direct excerpt from the docs source required, or is a paraphrase sufficient (with clear source backing either way)?
• What form of attribution is preferred (source title, URL, section heading, and/or excerpt)? Attribution is always required.
• If results are sparse after one refinement pass, should uncertainty be reported immediately?

Follow-Up Questions — EDS Research

Clarifying questions to sharpen the research scope before dispatching to agents/eds.agent.md. Pick the relevant section and skip questions already answered.

Scope

• Is this a question about a specific component's props, a usage example, accessibility guidance, or a design token?
• Is the component name already known, or does it need to be discovered first?
• Should the answer include a concrete code example, or is a prop table sufficient?
• Is there existing code that should be compared against the EDS recommendation?

Inputs

• Which EDS component or token family is involved, if known?
• Is there a specific prop, variant, or state that needs clarification?
• Is accessibility compliance (ARIA, keyboard, screen reader) part of the requirement?
• What level of depth is needed: quick prop lookup, full usage pattern, or accessibility deep-dive?

Evidence requirements

• Is a link to the EDS source required in the final answer, or is a paraphrase sufficient?
• Should the answer cover both the component API and a usage example, or just one?
• If results are weak after one refinement pass, should uncertainty be reported immediately, or is a related component fallback acceptable?

Follow-Up Questions — Framework Research

Clarifying questions to sharpen the research scope before dispatching to agents/framework.agent.md. Pick the relevant section and skip questions already answered.

Scope

• Is this a question about a specific hook, function, or class — or about a package/module boundary?
• Is there an exact symbol name already known, or does the symbol need to be discovered first?
• Should the answer include a concrete usage example, or is an API definition enough?
• Is the question about current framework behavior, or about verifying that existing code matches it?

Inputs

• Which Fusion Framework package or module is involved, if known?
• If a cookbook example is needed, is there existing code that should be compared against it?
• What level of depth is needed: quick lookup, package overview, or full cookbook example?
• Are there related symbols, hooks, or modules that should be cross-referenced?

Evidence requirements

• Is a reference to metadata.source required in the final answer, or is a paraphrase sufficient?
• Should the answer include a direct excerpt from the source, or just a summary?
• If results are weak after one refinement pass, should uncertainty be reported immediately, or is a fallback to docs acceptable?

Follow-Up Questions — Skills Research

Clarifying questions to sharpen the research scope before dispatching to agents/skills.agent.md. Pick the relevant section and skip questions already answered.

Scope

• Is this a capability fit question (does a skill for X exist?), a scope boundary question (what does skill Y cover?), a companion relationship question, or a usage guidance question?
• Is the skill name already known, or does it need to be discovered first?
• Is the user looking for a single skill or a combination of skills for a workflow?
• Is the question about what a skill does, or about what it explicitly does not do?

Inputs

• Which skill names are already known, if any?
• What workflow or capability is the user trying to cover?
• Are companion or orchestrator relationships part of what needs to be confirmed?
• Should the answer compare two or more skills, or focus on one?

Evidence requirements

• Should the source label (mcp_fusion_skills, local SKILL.md, or GitHub-backed catalog) be included in the final answer?
• Is a direct excerpt from the skill's description or metadata required, or is a summary sufficient?
• If the first MCP pass and the fallback both yield insufficient results, should uncertainty be reported immediately?

Source-Backed Answer Template

Use this structure when returning a Fusion research answer.

Scope

• Domain: framework / EDS / skills / redirected
• Agent used: agents/framework.agent.md / agents/eds.agent.md / agents/skills.agent.md
• Refinement used: yes / no

Answer

• Give the shortest answer that is actually supported by the evidence.

Evidence

• Source: <metadata.source or skill name or SKILL.md path>
• Why it matters: <what this source confirms>
• Excerpt: <the minimal excerpt or paraphrase needed>
• Source: <optional second source>
• Why it matters: <optional second angle>
• Excerpt: <the minimal excerpt or paraphrase needed>

Assumptions or uncertainty

• State what is confirmed.
• State what is inferred.
• If results stayed weak after one refinement pass, say that explicitly.

Next step

• If still in scope: name the next symbol, component, token, or skill to verify.
• If out of scope: redirect to fusion-app-react-dev (implementation), fusion-discover-skills (lifecycle), fusion-skill-authoring (create/edit), or fusion-mcp (MCP setup).

Changelog

0.4.1 - 2026-05-07

patch

• #170 `5e43223` - Apply caveman-compress prose style to references.

• Compress backend-code.query and framework.query references

0.4.0 - 2026-04-22

minor

• #143 `3c02e1d` Thanks @alftore! - Add backend-code routing support to fusion-research.

• Add backend-code domain classification and agent dispatch guidance
• Add backend-code research agent for C# and backend service implementation questions
• Add backend-code query reference with evidence-first search lanes
• Extend compatibility and MCP suggestions with backend code search support

0.3.0 - 2026-03-21

minor

• #110 `859fd0c` - Add docs domain to fusion-research skill

• Add agents/docs.agent.md routing platform documentation questions to mcp_fusion_search_docs
• Add assets/docs.follow-up.md for pre-dispatch scope questions
• Add references/docs.query.md with proven query lanes for docs retrieval
• Update SKILL.md description, classification table, dispatch table, agents list, and assets list to cover the new Docs domain
• Docs domain covers: Fusion platform concepts, onboarding, platform operations, and governance

Resolves equinor/fusion-core-tasks#411

0.2.0 - 2026-03-21

minor

• #106 `1f8a01d` - Rename fusion-research-framework to fusion-research and expand into a multi-domain research orchestrator.

• rename skill from fusion-research-framework to fusion-research
• expand scope from framework-only to three research domains: Framework, EDS, and skill catalog
• add agents/framework.agent.md — source-backed Fusion Framework API and cookbook research via mcp_fusion_search_framework
• add agents/eds.agent.md — source-backed EDS component props, usage, accessibility, and design token research via mcp_fusion_search_eds
• add agents/skills.agent.md — source-backed skill catalog lookup, scope boundary, and companion relationship research via mcp_fusion_skills
• add assets/framework.follow-up.md, assets/eds.follow-up.md, assets/skills.follow-up.md — pre-dispatch scope-sharpening questions per domain
• add references/framework.query.md, references/eds.query.md, references/skills.query.md — repeatable query patterns and evidence checklists per domain
• update assets/source-backed-answer-template.md to cover all three domains
• connect to fusion-mcp as next-step redirect when MCP is not yet running
• tighten discovery contract: description, trigger phrases, and "When not to use" boundaries
• apply council review fixes: remove erroneous cookbook lane reference from EDS agent, orphan follow-up assets linked from SKILL.md, mcp_fusion_search_docs added to suggested MCP list

Resolves equinor/fusion-core-tasks#837 Resolves equinor/fusion-core-tasks#838

patch

• #107 `d75d8c6` - Switch skills catalog agent to mcp_fusion_search_skills for source-backed retrieval.

• update agents/skills.agent.md to use mcp_fusion_search_skills (semantic search over local skills index) instead of the advisory mcp_fusion_skills
• update mcp.suggested list in metadata: replace mcp_fusion_skills with mcp_fusion_search_skills
• update compatibility line and agent description to reflect the correct tool
• update source labels in references/skills.query.md evidence checklist

Resolves equinor/fusion-core-tasks#834

0.1.0 - 2026-03-20

minor

• #103 `06b664d` - Add the fusion-research-framework skill under skills/.system/ for source-backed Fusion Framework implementation lookup through MCP.

• route framework research to the framework-focused MCP index
• define repeatable search lanes for APIs, package docs, and cookbook examples
• require source-path and excerpt evidence before finalizing answers
• limit refinement to one pass and redirect out-of-scope requests cleanly

resolves equinor/fusion-core-tasks#410

Backend Code Search Reference

Overview

C# implementation search across Fusion services, libraries, and infrastructure. Use for service implementations, API contracts, architectural patterns, and cross-service interactions.

Search lanes

Proven search starters

Start with high-level intent when service/method is unknown:

"How does fusion-core-services handle authorization for sensitive endpoints"
→ Returns authorization handlers, requirement definitions, Startup config

"What's the pattern for validating input in fusion services"
→ Returns FluentValidation classes, validator registrations, error handling

"How do Fusion services integrate with external APIs"
→ Returns HttpClient setup, API client interfaces, authentication patterns

"How is the People service structured"
→ Returns Controllers, CQRS handlers, Domain models, DbContext

Query construction tips

1. Use service names: "fusion-core-services", "fusion-libraries", "people", "org", "context" 2. Use architectural terms: "IRequest handler", "command", "query", "validator", "controller" 3. Use intent verbs: "how is X implemented", "what pattern does Y use", "does X support Z" 4. Add context: specific API version, scenario, or related entity helps narrow results

Evidence capture

Capture for backend code answers:

Weak signal indicators

State uncertainty if:

• Search returns no results for the service/method you're looking for
• Results are from unrelated services (e.g., Org service results when querying People service)
• Code snippets are incomplete (truncated methods, missing context)
• Results don't clearly answer the "how" or "why" part of the question after one refinement

Example: _"I found references to PeopleService but not the specific implementation of profile update logic. To complete this answer, I'd need to verify the UpdateProfile method locally."_

Lane examples by question type

"How does the authorization work?"

Lane: Authorization Pattern Query: "authorization requirement in fusion-core-services" + "role-based access" Expected result types:

• Authorization handler classes
• Requirement builder extensions
• Startup configuration showing authorization pipelines
• Example authorize attributes on controllers

"How do services call external APIs?"

Lane: Cross-Service Call Query: "how fusion services integrate with external API" + "HttpClient" OR "API client abstraction" Expected result types:

• API client interface definitions
• HttpClient factory setup
• Example service-to-service calls with URL construction
• Authentication header injection

"What's the pattern for validating user input?"

Lane: Validation Pattern Query: "FluentValidation" + "fusion-core-services" OR "validation rule" Expected result types:

• Validator classes inheriting from AbstractValidator<T>
• Validation pipeline registration in Startup
• Error response mappings
• Example rule chains

"How are domain events published?"

Lane: Event Handling Query: "domain event" + "publish" + "async messaging" OR "INotification" Expected result types:

• Domain event classes
• MediatR notification definitions
• Event handler implementations
• Service Bus or event bus integration code

Query Patterns

Use these repeatable search lanes when researching Fusion platform documentation through MCP.

Start with the smallest lane that fits

Proven examples

• Concept lookup: Fusion context model platform overview
• Onboarding: app developer onboarding Fusion setup access
• Operations: service discovery Fusion platform configuration environment
• Governance: access control Fusion platform guidelines governance

One refinement pass only

If the first pass is weak or ambiguous:

1. Rephrase using synonyms or alternate terminology. 2. Broaden the scope by dropping specific qualifiers and trying the parent concept. 3. Split compound questions into simpler sub-queries and run them separately. 4. Stop after the second pass and report uncertainty if the evidence is still weak.

Evidence checklist

Before using a result in the final answer, capture:

• metadata.source
• the excerpt that supports the point
• the platform domain or context (e.g., onboarding, access management, portal operations)

EDS Query Patterns

Use these repeatable query patterns when researching EDS through MCP.

Start with the pattern that fits

Proven examples

• Props: Button variant disabled onClick props
• Example: NativeSelect usage form label controlled
• Accessibility: Dialog accessibility focus trap keyboard
• Token: interactive primary color token eds

One refinement pass only

If the first pass is weak or ambiguous:

1. Add the component name explicitly. 2. Narrow to a specific prop, token, or accessibility concern. 3. Switch from props lookup to usage or accessibility pattern. 4. Stop after the second pass and report uncertainty if the evidence is still weak.

Evidence checklist

Before using a result in the final answer, capture:

• the source path
• the component or token name
• the minimal excerpt that supports the claim
• whether it is a prop definition, usage example, accessibility note, or design token

Prefer a second source only when it adds missing context, not just volume.

Query Patterns

Search lanes for researching Fusion Framework implementation via MCP.

Start with the smallest lane that fits

Proven examples

• API lookup: useFramework useFrameworkModule hook React framework context
• Package overview: FrameworkConfigurator init Fusion modules API
• Example lookup: router loader fusion.modules.http createHttpClient example

One refinement pass only

If the first pass is weak or ambiguous:

1. Add the exact symbol, package, or module name. 2. Narrow the type when the lane is still broad. 3. Switch lanes when needed:

• API lookup -> markdown for package context
• package overview -> tsdoc for an exact symbol
• example search -> cookbook

4. Stop after the second pass and report uncertainty if the evidence is still weak.

Evidence checklist

Capture before using a result:

• metadata.source
• metadata.type
• the excerpt that supports the point
• any useful package or symbol metadata from metadata.attributes

Important: cookbook results are indexed raw and contain actual source code — compare directly with user code when relevant. tsdoc and markdown results contain extracted docs only; note what still needs local verification.

Skills Query Patterns

Use these repeatable query patterns when researching the Fusion skill catalog through MCP.

Start with the intent that fits

Proven examples

• Capability fit: github pull request review skill
• Scope: fusion-skill-authoring when to use scope
• Companion: fusion-app-react-dev companion skills metadata

One refinement pass, then fallback

If the first pass is weak or ambiguous:

1. Add the exact skill name. 2. Try a narrower capability keyword. 3. Fall back to reading local SKILL.md files or GitHub-backed catalog content when MCP remains weak. 4. Stop after the fallback and report uncertainty if evidence is still insufficient.

Evidence checklist

Before using a result in the final answer, capture:

• the skill name
• the source label (mcp_fusion_search_skills, mcp_fusion_skills, local SKILL.md, or GitHub-backed catalog)
• the excerpt that supports the claim
• any companion, orchestrator, or overlap metadata when the question involves relationships

Prefer a second source only when it adds missing context — scope boundaries, companion relationships, or usage guidance not in the first result.