Provides various test operations for development and debugging. Supports echo operations, random data generation, error simulation, delay operations, and never-completing tasks. Perfect for testing async behavior, error handling, and orchestration workflows in the Covia platform.
| Name: | test |
| Class: | covia.adapter.TestAdapter |
This adapter provides 22 MCP tool(s) for AI agent integration:
| Tool Name | Description |
|---|---|
| test_taskllm | Test LLM that completes tasks. Reads task context from user messages, calls complete_task for the first task, then returns a text summary. Used for testing the full agent:request pipeline. |
| test_workspacellm | Test LLM that exercises workspace tools (covia_write, covia_append, covia_read) in sequence within a single agent run. Used for end-to-end testing of agent workspace interaction. |
| test_delay | Runs another op after a delay |
| test_selfchat | Test LLM that invokes agent:message on itself during its first transition, then handles the FOLLOWUP message on the next cycle. Used to regression-test nested self-chat during the agent run loop. |
| test_wakeresponse | Test transition that returns a wakeTime from state.wakeTime in the transition result. Used to exercise per-thread scheduler wire-up (B8.8). |
| test_compactllm | Test LLM that exercises the compact harness tool. First call returns tool calls for test:echo and compact; second call verifies the compacted segment and returns text. |
| test_random | Generates a specified number of random bytes using a cryptographically secure random number generator |
| test_badargsllm | Test LLM that emits a tool call with malformed (non-JSON) arguments on its first turn, then echoes the tool result content. Used for testing that broken tool arguments produce a visible tool error the LLM can react to, never a silent empty-map invocation. |
| test_chat | A multi-turn test operation that echoes messages back. Send messages via POST /api/v1/jobs/{id}. Send {"content": "done"} to complete. |
| test_nevertoolllm | Test LLM that calls v/test/ops/never as its first tool (blocking for toolCallTimeoutMs so tests can observe mid-run lattice state), then returns text after the timeout error. |
| test_subgoalllm | Test LLM driving subgoal recursion with an observable window: root issues a subgoal; the child calls v/test/ops/never (blocking), then completes; the root completes on the subgoal result. |
| test_neverfailllm | Test LLM that calls v/test/ops/never as its first tool, then FAILS the L3 call after the tool result — a scripted mid-cycle transition failure for lattice-frames durability tests. |
| test_error | Always fails, regardless of input |
| test_echo | Returns the input unchanged. Useful for testing or as a generalised identity function. |
| test_taskcomplete | Test transition function that auto-completes all tasks. Returns taskResults with each task's input as output. |
| test_never | An operation that never completes, i.e. its status will remain PENDING |
| test_pause | A test operation that immediately pauses itself. Send any message to unpause and complete the job with the original input as output. |
| test_toolllm | Test LLM that requests tool calls. On first call returns a tool call for test:echo; after tool results arrive returns a text summary. Used for testing the agent tool call loop. |
| test_llm | Test LLM operation that echoes the last user message. Accepts a messages array and returns an assistant message map. Used for testing the agent loop without a real LLM. |
| test_capturectx | Test-only transition op: records the RequestContext it ran under (keyed by agentId) and echoes the input. Used to assert agent run-loop identity (#91). |
| orchestrator | Runs a sequence of operations |
| test_loopllm | Test LLM that ALWAYS requests a tool call and never returns text or completes a task. Drives the agent tool-call loop to its iteration limit, to verify the agent fails the Job on give-up rather than leaving a task STARTED (covia-ai/covia#138). |