Skip to content

Operators & Planes

Every CTX statement begins with an operator-plane pair: the operator says what to do, the plane says where to do it.

Semantic or fuzzy search across a plane. Returns ranked results.

?k "authentication patterns" #code @30d ^5
?t github
?m #architecture @7d ^10
?s "process 3D assets" ^3

Use ? when you don’t know the exact name of what you’re looking for. The gateway performs semantic matching (knowledge plane) or tool-name fuzzy search (tools plane).

Exact lookup by ID, name, or reference. Returns one result or an error.

!t godot.assets.decimate_mesh
!k doc-abc123
!m auth-decision
!s godot-mcp-roles

Use ! when you know exactly what you want. Faster than ? because it skips ranking.

Execute a tool or skill. Returns structured output from the backend.

>t github.issues.create title="Fix SSE bug" labels="bug,p1"
>t godot.assets.decimate_mesh sourcePath="res://dragon.glb" targetRatio=0.1
>s deploy-pipeline env="staging"
>l "summarize this context" model="claude-sonnet"

Only valid on executable planes: tools (t), skills (s), and llm (l). Blocked on data planes (k, m) — you can’t “call” a memory.

Write new data to a plane.

+m auth-decision #architecture "JWT for API, session tokens for WebSocket"
+k "documentation" source="docs/api.md"
+t custom-tool endpoint="https://api.example.com/v1"
+s new-workflow "Steps: lint, test, deploy"
+i grammar

Modify existing data in a plane.

~m auth-decision "Updated: switched to PASETO tokens"
~k doc-abc123 "Revised content with new API endpoints"

Requires a target that already exists. If it doesn’t, use + instead.

Remove data from a plane.

-m old-decision
-k obsolete-doc
-t deprecated-tool

Delegate work to another agent. Only valid on the agents (a) plane.

^a code-review agent="claude-opus" context="PR #42"

The tools plane provides MCP (Model Context Protocol) tool discovery, inspection, and execution.

?t github Search for tools matching "github"
!t github.issues.list Get the schema for a specific tool
>t github.issues.create title="New issue" Call a tool

Tools are scoped by roles configured in actx.yaml. An agent with roles: [code] won’t see tools from a backend with roles: [admin].

The knowledge plane provides semantic search across ingested documents.

?k "authentication" #code @7d ^3 Search with tag + time filter + limit
+k "api-reference" source="docs/" Ingest a document
!k doc-abc123 Lookup by document ID

Knowledge sources are defined in actx.yaml:

planes:
knowledge:
enabled: true
sources: ["docs/", "src/"]

The memory plane provides persistent, cross-session storage with a 5-layer architecture:

LayerTagScope
Ephemeral#ephemeralSingle session only
Individual#myPer-agent, cross-session
Team#teamShared within a team
Organization#orgOrg-wide knowledge
Global#globalPlatform-wide facts
+m "lesson" #my "Singleton pattern breaks parallel tests"
?m #team @30d ^10
~m lesson-42 "Updated: use factory pattern instead"
?m @similar "auth-decision" #k:5

Memory also supports kind tags for semantic classification:

KindUse
#factVerified facts
#lessonLearned insights
#preferenceAgent/user preferences
#spatialLocation or layout data
#temporalTime-bound information
#socialRelationship data
#relationGraph relationships
#handoffInter-agent state transfer

The skills plane manages reusable agent workflows.

?s "process 3D assets" ^3 Find relevant skills
!s godot-mcp-roles Inspect a specific skill
>s deploy-pipeline env="staging" Execute a skill workflow

Skills are loaded from .md files in directories configured as skill sources.

The agents plane enables multi-agent coordination.

?a "code reviewer" Discover available agents
!a agent-123 Lookup a specific agent
^a code-review agent="claude-opus" context="PR #42" Delegate work
-a agent-123 Unregister an agent

The inspection plane provides runtime introspection of the CTX system itself.

?i grammar Get the full operator × plane matrix and syntax rules
+i grammar Write inspection data

Agents can query the CTX grammar at runtime without needing documentation:

?i grammar
// Returns: operators, planes, valid combinations, filter syntax, examples

The LLM plane provides direct access to language model inference.

>l "Summarize this context" model="claude-sonnet"
?l "embedding" input="mesh decimation algorithm"