# #26822 runtime transcript (synthetic data, fresh dev DB, T3CODE_DEV_SEED=0)
Environment: T3 dev server (`vp run dev`) on Linux, web client, Pi 1.4.0, model anthropic/claude-opus-5-5, full access.
Thread A (Pi): "Remember this codeword for later: PINEAPPLE-7731. Reply with only "noted"."

## Before: main dac8b11ddc
Prompt in thread B: "Call the tool mcp__t3_code__t3_thread_read directly (do not search for tools first) with threadId "<thread A>", limit 40, maxCharsPerItem 1500. Then tell me the codeword ..."
Result: `Tool mcp__t3_code__t3_thread_read not found`
Variant: the same call with `mcp__t3-code__t3_thread_read` and no search -> `Tool mcp__t3-code__t3_thread_read not found` (Pi deferred design; out of scope).
Variant: "Read T3 thread <thread A> and tell me the codeword" -> tool_search, then t3_thread_read -> PINEAPPLE-7731.

## After: fix/issue-26822 b51ded26f3
Same prompt, new thread -> the single tool call was `mcp__t3_code__t3_thread_read` -> "The codeword in that thread is PINEAPPLE-7731."
Repeated with a before_provider_request logger loaded via `pi -e` (instance launch argument):
every provider request declared 9 tools (read, bash, edit, write, tool_search, Pi's deferred placeholder, the 3 direct mcp__t3-code__ tools) and 0 mcp__t3_code__ names. The call succeeded again.

## After: fix/issue-26822 eb5723f4c9 (aliases `model-only`, following their public tool)
New thread, prompt: call mcp__t3_code__t3_thread_read directly (no search) on thread A, then mcp__t3_code__orchestrator_capabilities directly.
Pi session record: call mcp__t3_code__t3_thread_read -> isError false (thread A payload); call mcp__t3_code__orchestrator_capabilities -> isError false.
Request logger: 2 provider requests, each declaring 9 tools, 0 mcp__t3_code__ names (public direct tools: orchestrator_capabilities, delegate_task, task_status).