Build your test plan
Evaluation planner
Define one action, choose evaluation checks, and download a test plan with docs and public source links.
For developers
Choose the contract your integration needs. A Python policy check, an executable decision receipt, and a messaging response establish different things.
Explore the kernel's four verdicts. ALLOW and TRANSFORM can authorize a verified call. DENY and ESCALATE stop execution.
Set which actions, resources, and conditions are allowed for a governed tool.
Evaluate the proposed action and issue its decision receipt. A transformed call uses only the approved arguments.
Choose a decision in the record to see where execution stops.
Check the receipt against the exact call before running the tool. Retain the decision and audit record for review.
Illustrative decision
Send a report to a client
Decision
Result
Escalate · Execution stopped
The decision is recorded. Its receipt does not authorize execution.
Illustrative policy outcomes. No signed receipts are generated and no tools are executed.
Your evaluation starting point
Product-specific setup instructions, integration references, and source evidence.
Follow the product-specific setup and inspect the expected result before expanding the integration.
Open the documentationChoose an evaluation
Begin with a single tool call, proposed Python action, or cloud plan. Keep the execution path and evidence configuration in scope.
Build your test plan
Define one action, choose evaluation checks, and download a test plan with docs and public source links.
Tool execution
Require a valid decision receipt before an integrated tool call can run.
Only tool paths routed through the governed executor are covered.
Find your framework
Connect ACGS-Lite 3.0.1 rule evaluation to FastMCP, LangChain, OpenAI Agents, Claude Agent SDK, CrewAI, Pydantic AI, Google ADK, Agno, Strands, Haystack, BeeAI, AG2, and Microsoft Agent Framework.
Python policy checks
Start with configurable policy checks and governed Python wrappers.
A successful validation alone does not authorize an external action. Inspect the separate execution gate and audit configuration.
Cloud-plan checks
Check Google Cloud Terraform plan JSON and retain control evidence.
Your pipeline enforces the result before apply.
Connect the workflow
Inspect the responsibilities of each component and the hooks your host actually invokes.
Multi-agent coordination
Coordinate task ownership, peer validation, and governance evidence.
Self-hosted messaging
Operate governed messaging and workflow endpoints as a separate service.
A bus acceptance response is not approval to execute a tool call.
Integration reference
Inspect pre-tool checks, output screening, final-response holds, and optional evidence capture.
Connect middleware hooks to the actual runtime. The reference panel uses fixture traces; its replay and export buttons do not call a backend.
Inspect and learn
Use the reference material to define authority, investigate risks, and review the evidence your integration should retain.
Architecture & bounded contexts
Workspace topology across 18 repositories, 10 canonical legitimacy concepts, SQLite EffectStore lifecycle, TLA+ models, and persistence boundaries.
Canonical glossary & URN identities
Normative definitions and versioned URN identities from acgs-lite/GLOSSARY.md (5a012c5) and wiki/concepts.md.
Formal TLA+ verification & Receipt v2
TLC-checked TLA+ models (FENCEABLE, BOUNDED, STAGEABLE), invariants I1–I3, Acgs-Swarm specs, and staged ACGS-Lite 4.0.0 Receipt v2.
Persistence & data lifecycles
In-memory defaults vs. SQLite, JSONL, and PostgreSQL durability boundaries across ACGS-Lite, Swarm, Agent Bus, and Enterprise AI Manager.
Framework and task brief
Copy a governance framework and a blank task brief for agent workflows.
Source-backed answers
Search source-backed answers about receipts, signatures, audit evidence, and product scope.
Risks and warning signs
A practical catalogue of agent-workflow risks, warning signs, and integration controls.
Operating arrangements
Choose operating modes and controls appropriate to an agent task.
Source, release, and CI review
A dated source, release, CI, and limitations index for the product family.
Choose one tool. Define what it may do. Inspect what happens when a request crosses the boundary.