feat(tui): wire normal text input to LLM via A2A facade #11230

New issue

Closed

opened 2026-05-15 14:15:13 +00:00 by hamza.khyari · 0 comments

hamza.khyari commented

2026-05-15 14:15:13 +00:00

Member

Metadata

Commit Message: feat(tui): wire normal text input to LLM via A2A facade
Branch: feature/m8-tui-llm-dispatch

Background and Context

The TUI shell is built and running but the normal text input path is not wired to the LLM. When a user types a message and presses Enter in NORMAL mode, the handler (on_input_submitted in tui/app.py:204–210) simply echoes the expanded text back into the conversation widget and stops. No A2A dispatch, no SessionWorkflow.tell(), no LLM call is made.

The A2A facade (a2a/facade.py) and SessionWorkflow.tell() are fully implemented and already used by the CLI (cli/commands/session.py). This is a pure wiring gap — no new infrastructure is needed.

Current Behaviour

User types a message in the TUI → text is echoed back in the conversation widget → nothing sent to an LLM.

# tui/app.py:204–210 — current dead end
preview = result.expanded_text
conversation.update(preview)   # ← just echoes the text, no LLM call

Expected Behaviour

User types a message → message is sent to the LLM via A2aLocalFacade.dispatch() → assistant response is rendered in the conversation widget → UI remains responsive during the LLM call.

Acceptance Criteria

Typing a message in NORMAL mode and pressing Enter dispatches the message to the LLM via A2aLocalFacade.dispatch(A2aRequest(method="message/send", ...))
The assistant response is displayed in the conversation widget after the LLM returns
The Textual event loop is not blocked during the LLM call (worker thread used)
If no actor is configured for the session, a user-friendly error is shown instead of an unhandled exception
A real database-backed session is created at TUI startup (not hardcoded "default")
All existing TUI behaviour (slash commands, shell exec, @ references) is unchanged

Supporting Information

Five concrete implementation steps:

Real session creation — replace hardcoded session_id="default" (app.py:108) with SessionService.create() at startup in run_tui() (commands.py)
Inject A2aLocalFacade — build and register a fully-wired facade in run_tui() (register session_service + session_workflow + provider_registry), pass it into CleverAgentsTuiApp.__init__
Dispatch NORMAL mode input — replace dead-end conversation.update(preview) with facade.dispatch(A2aRequest(method="message/send", params={"session_id": ..., "message": ...})) and render response.result["assistant_message"]
Handle SessionActorNotConfiguredError — display friendly message: "No actor configured. Use /persona set <name> or select an actor first."
Async worker — wrap the blocking facade.dispatch() call in self.run_worker() to prevent freezing the Textual event loop

Key files:

src/cleveragents/tui/app.py — on_input_submitted, __init__
src/cleveragents/tui/commands.py — run_tui()
src/cleveragents/a2a/cli_bootstrap.py — get_facade()
src/cleveragents/a2a/facade.py — A2aLocalFacade._handle_message_send()
src/cleveragents/application/services/session_workflow.py — SessionWorkflow.tell()

Subtasks

Create real session via SessionService.create() in run_tui() and pass session_id to CleverAgentsTuiApp
Build and wire A2aLocalFacade in run_tui() and inject into CleverAgentsTuiApp
Replace dead-end NORMAL mode handler with facade.dispatch() call and response rendering
Catch SessionActorNotConfiguredError and display user-friendly message
Wrap LLM dispatch in self.run_worker() for non-blocking execution
Tests (Behave): add scenarios covering the normal message flow and error paths
Verify coverage >= 97% via nox -s coverage_report
Run nox (all default sessions), fix any errors

Definition of Done

This issue is complete when:

All subtasks above are completed and checked off.
A Git commit is created where the first line matches the Commit Message in Metadata exactly.
The commit is pushed to the branch matching the Branch in Metadata exactly.
The commit is submitted as a PR to master, reviewed, and merged.

## Metadata - **Commit Message**: `feat(tui): wire normal text input to LLM via A2A facade` - **Branch**: `feature/m8-tui-llm-dispatch` ## Background and Context The TUI shell is built and running but the normal text input path is not wired to the LLM. When a user types a message and presses Enter in `NORMAL` mode, the handler (`on_input_submitted` in `tui/app.py:204–210`) simply echoes the expanded text back into the conversation widget and stops. No A2A dispatch, no `SessionWorkflow.tell()`, no LLM call is made. The A2A facade (`a2a/facade.py`) and `SessionWorkflow.tell()` are **fully implemented** and already used by the CLI (`cli/commands/session.py`). This is a pure wiring gap — no new infrastructure is needed. ## Current Behaviour User types a message in the TUI → text is echoed back in the conversation widget → nothing sent to an LLM. ```python # tui/app.py:204–210 — current dead end preview = result.expanded_text conversation.update(preview) # ← just echoes the text, no LLM call ``` ## Expected Behaviour User types a message → message is sent to the LLM via `A2aLocalFacade.dispatch()` → assistant response is rendered in the conversation widget → UI remains responsive during the LLM call. ## Acceptance Criteria - Typing a message in NORMAL mode and pressing Enter dispatches the message to the LLM via `A2aLocalFacade.dispatch(A2aRequest(method="message/send", ...))` - The assistant response is displayed in the conversation widget after the LLM returns - The Textual event loop is not blocked during the LLM call (worker thread used) - If no actor is configured for the session, a user-friendly error is shown instead of an unhandled exception - A real database-backed session is created at TUI startup (not hardcoded `"default"`) - All existing TUI behaviour (slash commands, shell exec, `@` references) is unchanged ## Supporting Information Five concrete implementation steps: 1. **Real session creation** — replace hardcoded `session_id="default"` (`app.py:108`) with `SessionService.create()` at startup in `run_tui()` (`commands.py`) 2. **Inject `A2aLocalFacade`** — build and register a fully-wired facade in `run_tui()` (register `session_service` + `session_workflow` + `provider_registry`), pass it into `CleverAgentsTuiApp.__init__` 3. **Dispatch NORMAL mode input** — replace dead-end `conversation.update(preview)` with `facade.dispatch(A2aRequest(method="message/send", params={"session_id": ..., "message": ...}))` and render `response.result["assistant_message"]` 4. **Handle `SessionActorNotConfiguredError`** — display friendly message: "No actor configured. Use `/persona set <name>` or select an actor first." 5. **Async worker** — wrap the blocking `facade.dispatch()` call in `self.run_worker()` to prevent freezing the Textual event loop Key files: - `src/cleveragents/tui/app.py` — `on_input_submitted`, `__init__` - `src/cleveragents/tui/commands.py` — `run_tui()` - `src/cleveragents/a2a/cli_bootstrap.py` — `get_facade()` - `src/cleveragents/a2a/facade.py` — `A2aLocalFacade._handle_message_send()` - `src/cleveragents/application/services/session_workflow.py` — `SessionWorkflow.tell()` ## Subtasks - [ ] Create real session via `SessionService.create()` in `run_tui()` and pass `session_id` to `CleverAgentsTuiApp` - [ ] Build and wire `A2aLocalFacade` in `run_tui()` and inject into `CleverAgentsTuiApp` - [ ] Replace dead-end NORMAL mode handler with `facade.dispatch()` call and response rendering - [ ] Catch `SessionActorNotConfiguredError` and display user-friendly message - [ ] Wrap LLM dispatch in `self.run_worker()` for non-blocking execution - [ ] Tests (Behave): add scenarios covering the normal message flow and error paths - [ ] Verify coverage >= 97% via `nox -s coverage_report` - [ ] Run `nox` (all default sessions), fix any errors ## Definition of Done This issue is complete when: - All subtasks above are completed and checked off. - A Git commit is created where the first line matches the Commit Message in Metadata exactly. - The commit is pushed to the branch matching the Branch in Metadata exactly. - The commit is submitted as a PR to master, reviewed, and merged.

Rows
Columns