tool_search
What you are trying to do, in plain language. e.g.
T1 Primitives
Canonical agentic primitives — the Layer-2 chatmode catalog. Every chat turn ships these 16 meta-tools.
Discovery
Discovery primitives — find tools/agents at runtime
agent_search
What kind of agent you need, in plain language. e.g.
memory_search
Search the user\
Sub-agent lifecycle
Spawn, send, list, stop sub-agents
Task
Azure IAM audit
agent_send
The agent_session_id of the running sub-agent (from agent_list or the original Task tool_use_id).
agent_list
Enumerate the live (running) sub-agent sessions tied to this chat.
agent_stop
The agent_session_id of the running sub-agent to terminate (from agent_list).
IO
Read/write the outside world — web, large results, synth tools
read_large_result
Retrieve a paged slice of a previously-stored large tool result by its artifactHandle.
web_search
Plain-language search query. Examples:
web_fetch
Absolute URL to fetch. Must be http:// or https://. Server enforces SSRF denylist (no IMDS, no metadata endpoints, no RFC1918, no *.svc.cluster.local).
Memory
Save and recall exemplars across sessions
pattern_save
The original user request, verbatim or near-verbatim.
pattern_recall
The natural language query (usually the user prompt or your reformulation of it).
Visualization
Compose visuals, apps, and artifacts
compose_visual
Pick the template. Each has a strict data shape.
compose_app
Freestyle path — full HTML document including <!doctype html> +
render_artifact
The artifact kind.
Clarification
Ask the user when intent is ambiguous
request_clarification
The clarifying question to ask the user. Be specific and
Other
Other T1 primitives
includetasktool
T1 primitive: includeTaskTool
rag
T1 primitive: RAG
both
T1 primitive: BOTH
t1
T1 primitive: T1
not
T1 primitive: NOT
and
T1 primitive: AND
memorize
Short stable identifier for the memory (e.g.
generate_image
Detailed natural-language description of the image to generate.