UAT: `RuntimeExecuteActor.execute()` records stub invocations instead of dispatching real tool calls based on decision content #3819

New issue

Open

opened 2026-04-06 06:43:46 +00:00 by freemo · 0 comments

freemo commented

2026-04-06 06:43:46 +00:00

Owner

Metadata

Branch: fix/runtime-execute-actor-real-tool-dispatch
Commit Message: fix(executor): implement real tool dispatch in RuntimeExecuteActor
Milestone: (backlog — see note below)
Parent Epic: #368

Backlog note: This issue was discovered during autonomous operation
on milestone v3.4.0. It does not block milestone completion and has been
placed in the backlog for human review and future milestone assignment.

Summary

RuntimeExecuteActor.execute() in src/cleveragents/application/services/plan_execution_context.py (lines 328-442) records stub tool invocations for every decision node instead of actually dispatching real tool calls based on the decision content. This means the "runtime" execute actor is functionally equivalent to the stub actor — it never actually executes any tools.

What Was Tested

Code-level analysis of RuntimeExecuteActor.execute() against the specification §19513-19551 (Tool-Based Resource Modification) and §19205-19226 (Execute Phase).

Expected Behavior (from spec)

Per specification §19513-19551:

The Execute phase should dispatch real tool calls based on the decision tree
LLMs call tools directly (edit_file(), write_file(), etc.)
Tools operate on the sandbox
ChangeSet is built from actual tool invocations

Per docs/reference/plan_execute.md:

RuntimeExecuteActor wraps ToolRunner to execute strategy decisions with changeset capture
In later milestones, this will dispatch real tool calls based on the decision content

Actual Behavior

In plan_execution_context.py lines 381-404, for each decision, the actor:

Calls self._tool_runner.discover() (just lists available tools, doesn't execute any)
Creates a ToolInvocation with tool_name="stub/execute-step" — a hardcoded stub name
Records a ChangeEntry with operation=ChangeOperation.MODIFY and path=f"decisions/{decision_id}" — a synthetic path, not a real resource path

No actual tool is ever executed. The ToolRunner.execute() method is never called. The changeset contains only synthetic stub entries.

Code Location

src/cleveragents/application/services/plan_execution_context.py, lines 381-404
The comment on line 335 explicitly states: "For each decision, discovers available tools and records a stub invocation entry for each decision"
Line 388: tool_name="stub/execute-step" — hardcoded stub name

Impact

Plans in runtime mode (with execution_context set) appear to execute successfully but no actual tools are invoked. The changeset produced contains only synthetic entries. This means:

No real file modifications occur during Execute phase
The sandbox is never actually populated with work
Apply phase has nothing real to apply

Steps to Reproduce

Create a PlanExecutionContext with a real ToolRunner
Create a RuntimeExecuteActor with the context
Call actor.execute(decisions) with real decisions
Observe that ToolRunner.execute() is never called — only ToolRunner.discover() is called
Observe that the changeset contains only stub/execute-step entries

Subtasks

Implement real tool dispatch in RuntimeExecuteActor.execute() based on decision content
Map decision step_text to appropriate tool calls
Ensure ToolRunner.execute() is called for each tool invocation
Record real ChangeEntry objects from actual tool outputs
Add Behave tests for real tool dispatch behavior
Run nox (all default sessions), fix any errors
Verify coverage >= 97% via nox -s coverage_report

Definition of Done

This issue is complete when:

All subtasks above are completed and checked off.
RuntimeExecuteActor.execute() dispatches real tool calls via ToolRunner.execute() based on decision content
Changeset contains real change entries from actual tool invocations
All existing tests pass
New Behave tests cover the real dispatch behavior
A Git commit is created where the first line of the commit message matches the Commit Message in Metadata exactly, followed by a blank line, then additional lines providing relevant details about the implementation.
The commit is pushed to the remote on the branch matching the Branch in Metadata exactly.
The commit is submitted as a pull request to master, reviewed, and merged before this issue is marked done.
All nox stages pass
Coverage >= 97%

Automated by CleverAgents Bot
Supervisor: UAT Testing | Agent: ca-new-issue-creator

## Metadata - **Branch**: `fix/runtime-execute-actor-real-tool-dispatch` - **Commit Message**: `fix(executor): implement real tool dispatch in RuntimeExecuteActor` - **Milestone**: *(backlog — see note below)* - **Parent Epic**: #368 > **Backlog note:** This issue was discovered during autonomous operation > on milestone v3.4.0. It does not block milestone completion and has been > placed in the backlog for human review and future milestone assignment. ## Summary `RuntimeExecuteActor.execute()` in `src/cleveragents/application/services/plan_execution_context.py` (lines 328-442) records stub tool invocations for every decision node instead of actually dispatching real tool calls based on the decision content. This means the "runtime" execute actor is functionally equivalent to the stub actor — it never actually executes any tools. ## What Was Tested Code-level analysis of `RuntimeExecuteActor.execute()` against the specification §19513-19551 (Tool-Based Resource Modification) and §19205-19226 (Execute Phase). ## Expected Behavior (from spec) Per specification §19513-19551: - The Execute phase should dispatch real tool calls based on the decision tree - LLMs call tools directly (`edit_file()`, `write_file()`, etc.) - Tools operate on the sandbox - ChangeSet is built from actual tool invocations Per `docs/reference/plan_execute.md`: - `RuntimeExecuteActor` wraps `ToolRunner` to execute strategy decisions with changeset capture - In later milestones, this will dispatch real tool calls based on the decision content ## Actual Behavior In `plan_execution_context.py` lines 381-404, for each decision, the actor: 1. Calls `self._tool_runner.discover()` (just lists available tools, doesn't execute any) 2. Creates a `ToolInvocation` with `tool_name="stub/execute-step"` — a hardcoded stub name 3. Records a `ChangeEntry` with `operation=ChangeOperation.MODIFY` and `path=f"decisions/{decision_id}"` — a synthetic path, not a real resource path No actual tool is ever executed. The `ToolRunner.execute()` method is never called. The changeset contains only synthetic stub entries. ## Code Location - `src/cleveragents/application/services/plan_execution_context.py`, lines 381-404 - The comment on line 335 explicitly states: "For each decision, discovers available tools and records a stub invocation entry for each decision" - Line 388: `tool_name="stub/execute-step"` — hardcoded stub name ## Impact Plans in runtime mode (with `execution_context` set) appear to execute successfully but no actual tools are invoked. The changeset produced contains only synthetic entries. This means: - No real file modifications occur during Execute phase - The sandbox is never actually populated with work - Apply phase has nothing real to apply ## Steps to Reproduce 1. Create a `PlanExecutionContext` with a real `ToolRunner` 2. Create a `RuntimeExecuteActor` with the context 3. Call `actor.execute(decisions)` with real decisions 4. Observe that `ToolRunner.execute()` is never called — only `ToolRunner.discover()` is called 5. Observe that the changeset contains only `stub/execute-step` entries ## Subtasks - [ ] Implement real tool dispatch in `RuntimeExecuteActor.execute()` based on decision content - [ ] Map decision `step_text` to appropriate tool calls - [ ] Ensure `ToolRunner.execute()` is called for each tool invocation - [ ] Record real `ChangeEntry` objects from actual tool outputs - [ ] Add Behave tests for real tool dispatch behavior - [ ] Run `nox` (all default sessions), fix any errors - [ ] Verify coverage >= 97% via `nox -s coverage_report` ## Definition of Done This issue is complete when: - All subtasks above are completed and checked off. - `RuntimeExecuteActor.execute()` dispatches real tool calls via `ToolRunner.execute()` based on decision content - Changeset contains real change entries from actual tool invocations - All existing tests pass - New Behave tests cover the real dispatch behavior - A Git commit is created where the **first line** of the commit message matches the Commit Message in Metadata exactly, followed by a blank line, then additional lines providing relevant details about the implementation. - The commit is pushed to the remote on the branch matching the **Branch** in Metadata exactly. - The commit is submitted as a **pull request** to `master`, reviewed, and **merged** before this issue is marked done. - All nox stages pass - Coverage >= 97% --- **Automated by CleverAgents Bot** Supervisor: UAT Testing | Agent: ca-new-issue-creator

freemo added a new dependency

2026-04-06 06:44:00 +00:00

#368 Epic: Subplans & Parallelism

freemo added the

Points

label

2026-04-06 06:50:03 +00:00

freemo referenced this issue

2026-04-06 06:54:29 +00:00

[Automated] CleverAgents Build Session - 2026-04-06 #3775

freemo referenced this issue

2026-04-06 07:04:26 +00:00