UAT: Semantic Escalation controller is never invoked to gate automation #6993

New issue

Open

opened 2026-04-10 06:12:29 +00:00 by HAL9000 · 0 comments

HAL9000 commented

2026-04-10 06:12:29 +00:00

Owner

What was tested

Reviewed Semantic Escalation and automation profile sections (docs/specification.md lines 28483-28690) on commit 51aab18411.

Expected behavior

The runtime should feed plan decisions through the Semantic Escalation engine so each automatable flag compares the computed confidence score against the profile threshold.
Thresholds like edit_code, execute_command, create_file, install_dependency, modify_config, and approve_plan should pause for human input when confidence is below the configured value.

Actual behavior

AutonomyController.should_proceed_automatically(...) is never called outside of tests/benchmarks. No production module references it, so confidence-based gating never runs.
Plan lifecycle only checks raw threshold values for phase transitions (create_tool, select_tool, access_network, delete_content). All other flags are ignored, so plans proceed automatically regardless of confidence.

Steps to reproduce

From the repo root, run:

python3 - <<'PY'
import pathlib
root = pathlib.Path('src/cleveragents')
uses = [str(p.relative_to(root)) for p in root.rglob('*.py')
        if p.name != 'autonomy_controller.py' and 'should_proceed_automatically(' in p.read_text(encoding='utf-8')]
print('Semantic escalation uses:', uses)
PY

The script prints ['domain/models/core/escalation.py'], meaning only the domain model references the method; runtime code never does.
Inspect src/cleveragents/application/services/plan_lifecycle_service.py — automation thresholds are hard-coded checks against 1.0 and do not call the controller.

Code location

src/cleveragents/application/services/autonomy_controller.py (unused controller)
src/cleveragents/application/services/plan_lifecycle_service.py
src/cleveragents/application/services/plan_executor.py

## What was tested - Reviewed Semantic Escalation and automation profile sections (docs/specification.md lines 28483-28690) on commit 51aab184112728471a44d5a91c334663cf8cd016. ## Expected behavior - The runtime should feed plan decisions through the Semantic Escalation engine so each automatable flag compares the computed confidence score against the profile threshold. - Thresholds like `edit_code`, `execute_command`, `create_file`, `install_dependency`, `modify_config`, and `approve_plan` should pause for human input when confidence is below the configured value. ## Actual behavior - `AutonomyController.should_proceed_automatically(...)` is never called outside of tests/benchmarks. No production module references it, so confidence-based gating never runs. - Plan lifecycle only checks raw threshold values for phase transitions (`create_tool`, `select_tool`, `access_network`, `delete_content`). All other flags are ignored, so plans proceed automatically regardless of confidence. ## Steps to reproduce 1. From the repo root, run: ```bash python3 - <<'PY' import pathlib root = pathlib.Path('src/cleveragents') uses = [str(p.relative_to(root)) for p in root.rglob('*.py') if p.name != 'autonomy_controller.py' and 'should_proceed_automatically(' in p.read_text(encoding='utf-8')] print('Semantic escalation uses:', uses) PY ``` 2. The script prints `['domain/models/core/escalation.py']`, meaning only the domain model references the method; runtime code never does. 3. Inspect `src/cleveragents/application/services/plan_lifecycle_service.py` — automation thresholds are hard-coded checks against 1.0 and do not call the controller. ## Code location - `src/cleveragents/application/services/autonomy_controller.py` (unused controller) - `src/cleveragents/application/services/plan_lifecycle_service.py` - `src/cleveragents/application/services/plan_executor.py`

Rows
Columns