AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

foundry-iq

Foundry IQ knowledge bases. WHEN: make local or Blob documents searchable; create/diagnose KBs; triage unsupported connectors or multi-source KB creation/reconfiguration (even with classic-search terms); connect existing KB to agents (including multi-source KBs); create/reuse a Search service; retrieve from an existing knowledge base with citations; no brand words needed. Read its procedure before query/target questions. NOT: other KB providers, repository-file search, classic Azure AI Search index/query/app work with no Foundry IQ KB or connector request, generic agent creation.

Original source
microsoft
License
MIT
Pinned revision
3f5b461b2314e01c400564051e92469114c5b68f

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read microsoft--foundry-iq at revision 3f5b461b2314e01c400564051e92469114c5b68f, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 45

references/intent-routing.md

Intent, handoffs and completion examples

Read for ambiguous/compound goals, unresolved search-mode choices, or uncertainty about completion. These examples guide interpretation, not keyword-first matching.

Intent and completion

Route by the user's end result, not the first connector/resource keyword. Within Foundry IQ, "make these documents searchable" or "ask questions over these documents" means a usable KB with validated retrieval. Default to Create KB, not a source-only operation. Words such as files, Blob, index or embeddings identify prerequisites; they do not downscope a broader goal.

Confirm the intended finish once in normal intake, in user terms: "I'll make these documents searchable through a knowledge base and check retrieval against them. Do you want that full outcome, or only the knowledge source?" Do not re-ask an already explicit or confirmed goal. Use source-only only when the user explicitly requests or confirms that narrower outcome; never infer it from an intermediate step. Goal confirmation is not approval for Azure writes.

Retain the confirmed goal, selected data, explicit exclusions and remaining steps across handoffs. A child's completed result is not overall completion. For a KB goal, resume the KB owner after Search/model/source prerequisites with returned IDs and fresh source/readiness evidence. Present and approve the concrete KB plan, create/reuse and read back the KB, then run supported-question and unrelated-question retrieval checks under the selected API/output mode. Require source-backed results and original-source citations; source ingestion, KB existence or HTTP success alone is not validated retrieval.

Finish only when every requested outcome is verified. Otherwise report completed prerequisites, the blocker and remaining steps; never mark the whole goal complete or silently downgrade it to KS-only. Resume after an approved prerequisite rather than asking the user to restart. Never preapprove an unknown KB mutation. Explicit source-only, infrastructure-only and read-only requests remain bounded; never create an unrequested KB/agent. Explicit classic Search work still routes out.

Search-mode choice

For new unresolved search-mode intent, recommend/preselect Hybrid (keyword + vector), with Keyword-only (disable vectors) as the explicit opt-out. This recommendation is not consent or a helper default. Honor existing/informed keyword-only choices and exact reuse unchanged: do not re-ask settled choices, silently retrofit vectors or choose/deploy a model. An explicit change to an existing source needs a separate supported plan/approval; read-only retrieval never changes its source.

Ask in the grouped intake: "I recommend Hybrid (keyword + vector) for keyword precision plus semantic/paraphrase matching. Would you like Hybrid, or Keyword-only without vectors?" Where selectable defaults are supported, preselect Hybrid, but require an explicit customer choice; silence or an unanswered field is not consent. Reuse supplied informed choices and group compatible unresolved decisions rather than adding separate confirmations for each disclosure.

Before plan approval explain:

| Choice | Benefit and tradeoff | |---|---| | Hybrid (recommended) | Combines keyword precision with semantic/paraphrase matching, including related wording without exact terms. Adds embedding charges, ingestion/query embedding processing where applicable, vector storage/capacity, and content processing/data movement to the selected embedding service. Disclose selected model, region, auth/network and applicable cost/data boundaries; no guaranteed relevance improvement. | | Keyword-only (opt-out) | Keeps keyword matching without source vectors or their embedding charges/processing/storage. May miss paraphrases with little term overlap. Search, extraction and independently requested chat/CU can still cost money; this is not a free or model-free promise. |

Keyword-only: no embedding discovery. Discover only missing dependencies for the confirmed hybrid choice, using the bounded scoped helper; select an account before listing its deployments. Known compatible model choices need exact readback, not rediscovery. Never auto-select a deployment, deploy models, call models or upload content during intake. Missing/unavailable hybrid dependencies require supported scoped discovery, concretely approved setup, or explicit customer keyword-only opt-out; no silent keyword fallback. Report readback/access/capacity uncertainty precisely; metadata is candidate evidence, not effective readiness. Model-free minimal+extractive routes do zero model/CU discovery; keyword-only does not suppress independently required chat/CU discovery.

Retrieval keyword+vector is independent of text-only versus multimodal extraction. Neither search choice selects CU, KB reasoning/synthesis or semantic reranker automatically. A text-only PDF can use hybrid; multimodal content can use keyword-only. Assess extraction/answer dependence separately without defaulting Minimal or selecting Standard merely because an image/table exists.

Keep helper inputs explicit: File vectorization: none versus azureOpenAI; Blob minimal-lexical versus minimal-vector, or standard-cu with independently selected optional embeddings. File/Blob JSON examples are explicit keyword-only choices, not UX defaults. Omitted/null embeddings never become vectors; a selected vector mode with missing model choices blocks, not downgrades. Preserve separate sample-content consent and actual resource/permission/CU/source/KB approvals; do not approve unknown future bodies. Approved source creation/ingestion is not validated hybrid retrieval; use the existing vector/hybrid verification contract with explicit query approval, not a billable intake probe.

Scenario examples

Read the whole request; broader goals and explicit limits take precedence over a mentioned child step.

| User request | Start | Required finish | |---|---|---| | Make these local documents searchable. | Create KB | Confirm KB outcome; File prerequisite, valid KB and validated retrieval. | | Make this Blob container searchable. | Create KB | Confirm KB outcome; Blob prerequisite, valid KB and validated retrieval. | | Create a Blob knowledge source, then a knowledge base and test it. | Create KB | Both resources and validated retrieval; do not stop at the KS. | | Use my existing Search service to make these files searchable. | Create KB | Reuse Search; finish the KB and retrieval, not just infrastructure. | | Create only a Blob knowledge source; no KB. | Blob family | Verified source only; no KB or agent creation. | | Set up only a Search service. | Search | Verified service only; no source or KB creation. | | Query this existing KB. | Retrieve | Read-only source-backed retrieval; no creation. | | Connect this KB to my existing Foundry agent. | Connect | Verified requested connection and agent retrieval; configuration alone is insufficient. | | Make OneLake and Blob searchable together. | Diagnose | State unsupported scope, point to documentation, and hand off to agent to take direct action; never substitute a Blob-only success. | | Build a classic Search index and query it from my app. | Outside this skill | Classic Search handoff; do not invent a KB requirement. |

References

Authorities: failure/conflict/uncertainty only.