cleveragents/cleveragents-core

Fork 3

fix(plan): wire sandbox_root into plan execute pipeline #1313

New issue

Open

opened 2026-04-02 11:45:03 +00:00 by brent.edwards · 2 comments

brent.edwards commented

2026-04-02 11:45:03 +00:00

Member

Metadata

Commit Message: fix(plan): wire sandbox_root into plan execute pipeline
Branch: bugfix/m1-plan-execute-sandbox-root

Background

During development of #1249 (updating m1_acceptance.robot with output validation), it was discovered that plan execute does not write LLM-generated files to the sandbox directory. The LLM is invoked and successfully generates file content (e.g., HELLO.md), but the files are never written to disk.

Root Cause

_get_plan_executor() in src/cleveragents/cli/commands/plan.py (approx. line 1267) does not pass sandbox_root to PlanExecutor. This causes the following chain of failure:

PlanExecutor.__init__() receives sandbox_root=None (default)
PlanExecutor._run_execute_with_stub() passes sandbox_root=None to LLMExecuteActor.execute()
LLMExecuteActor.execute() checks if sandbox_root is not None and not read_only: before calling _write_to_sandbox() — this guard evaluates to False
_write_to_sandbox() is never called, so files generated by the LLM are silently discarded

Key Code Locations

File	Location	Role
`src/cleveragents/cli/commands/plan.py`	`_get_plan_executor()` (~line 1267)	Missing `sandbox_root` parameter
`src/cleveragents/application/services/plan_executor.py`	`__init__()` (~line 293)	Receives `None` for `sandbox_root`
`src/cleveragents/application/services/plan_executor.py`	`_run_execute_with_stub()` (~line 726)	Passes `None` to actor
`src/cleveragents/application/services/llm_actors.py`	`execute()` (~line 354)	Guard skips `_write_to_sandbox()`
`src/cleveragents/application/services/llm_actors.py`	`_write_to_sandbox()` (~line 409)	The file writer that is never called

Secondary Issues

plan apply does not merge sandbox: _lifecycle_apply_with_id() only transitions plan state metadata. It never calls GitWorktreeSandbox.commit() to merge changes from a sandbox into the target repo. Even if files were written to a sandbox, they would not make it into the target repository.
No GitWorktreeSandbox created in the execute pipeline: The sandbox infrastructure (infrastructure/sandbox/git_worktree.py) is fully implemented and tested in isolation, but is never instantiated or wired into the CLI plan execute flow.

Impact

This bug means the core M1 milestone acceptance criterion — "Execute/apply runs via actor-based LLM path producing tool invocations and a ChangeSet" — is not fully operational in the CLI. The LLM runs and generates output, but the generated files are silently lost.

Acceptance Criteria

_get_plan_executor() resolves the project's git-checkout resource path and creates or obtains a sandbox directory
sandbox_root is passed to PlanExecutor and propagated to LLMExecuteActor
LLMExecuteActor._write_to_sandbox() is called when the LLM generates file content
After plan execute, generated files exist in the sandbox directory
plan diff shows the generated file changes
plan apply merges sandbox changes into the target repository with a git commit
Post-apply, the generated files exist in the target repository
The tdd_expected_fail tag is removed from the M1 acceptance E2E test and the test passes

Definition of Done

This issue is complete when:

All acceptance criteria above are satisfied.
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.
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.

## Metadata - **Commit Message**: `fix(plan): wire sandbox_root into plan execute pipeline` - **Branch**: `bugfix/m1-plan-execute-sandbox-root` ## Background During development of #1249 (updating `m1_acceptance.robot` with output validation), it was discovered that `plan execute` does **not** write LLM-generated files to the sandbox directory. The LLM is invoked and successfully generates file content (e.g., `HELLO.md`), but the files are never written to disk. ### Root Cause `_get_plan_executor()` in `src/cleveragents/cli/commands/plan.py` (approx. line 1267) does **not** pass `sandbox_root` to `PlanExecutor`. This causes the following chain of failure: 1. `PlanExecutor.__init__()` receives `sandbox_root=None` (default) 2. `PlanExecutor._run_execute_with_stub()` passes `sandbox_root=None` to `LLMExecuteActor.execute()` 3. `LLMExecuteActor.execute()` checks `if sandbox_root is not None and not read_only:` before calling `_write_to_sandbox()` — this guard evaluates to `False` 4. `_write_to_sandbox()` is **never called**, so files generated by the LLM are silently discarded ### Key Code Locations | File | Location | Role | |------|----------|------| | `src/cleveragents/cli/commands/plan.py` | `_get_plan_executor()` (~line 1267) | **Missing `sandbox_root` parameter** | | `src/cleveragents/application/services/plan_executor.py` | `__init__()` (~line 293) | Receives `None` for `sandbox_root` | | `src/cleveragents/application/services/plan_executor.py` | `_run_execute_with_stub()` (~line 726) | Passes `None` to actor | | `src/cleveragents/application/services/llm_actors.py` | `execute()` (~line 354) | Guard skips `_write_to_sandbox()` | | `src/cleveragents/application/services/llm_actors.py` | `_write_to_sandbox()` (~line 409) | The file writer that is never called | ### Secondary Issues 1. **`plan apply` does not merge sandbox:** `_lifecycle_apply_with_id()` only transitions plan state metadata. It never calls `GitWorktreeSandbox.commit()` to merge changes from a sandbox into the target repo. Even if files were written to a sandbox, they would not make it into the target repository. 2. **No `GitWorktreeSandbox` created in the execute pipeline:** The sandbox infrastructure (`infrastructure/sandbox/git_worktree.py`) is fully implemented and tested in isolation, but is never instantiated or wired into the CLI `plan execute` flow. ### Impact This bug means the core M1 milestone acceptance criterion — "Execute/apply runs via actor-based LLM path producing tool invocations and a ChangeSet" — is not fully operational in the CLI. The LLM runs and generates output, but the generated files are silently lost. ## Acceptance Criteria - [ ] `_get_plan_executor()` resolves the project's git-checkout resource path and creates or obtains a sandbox directory - [ ] `sandbox_root` is passed to `PlanExecutor` and propagated to `LLMExecuteActor` - [ ] `LLMExecuteActor._write_to_sandbox()` is called when the LLM generates file content - [ ] After `plan execute`, generated files exist in the sandbox directory - [ ] `plan diff` shows the generated file changes - [ ] `plan apply` merges sandbox changes into the target repository with a git commit - [ ] Post-apply, the generated files exist in the target repository - [ ] The `tdd_expected_fail` tag is removed from the M1 acceptance E2E test and the test passes ## Definition of Done This issue is complete when: - All acceptance criteria above are satisfied. - 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. - 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.

brent.edwards added the

labels

2026-04-02 11:45:19 +00:00

brent.edwards referenced this issue

2026-04-02 11:47:23 +00:00

test(e2e): update m1_acceptance.robot #1260

brent.edwards referenced this issue

2026-04-02 11:47:55 +00:00

test(e2e): update m1_acceptance.robot #1260

freemo commented

2026-04-02 16:48:48 +00:00

Owner

Triage: Verified

This is a well-documented, critical bug filed by @brent.edwards during development of #1249 (M1 acceptance E2E test hardening). The root cause analysis is thorough and the code path is clearly traced.

Assessment:

Priority: Critical — confirmed. This bug means the core plan execute pipeline silently discards LLM-generated files, which breaks the M1 acceptance criterion ("Execute/apply runs via actor-based LLM path producing tool invocations and a ChangeSet").
Milestone: Assigning to v3.7.0. While this is an M1-era bug, it was discovered during v3.7.0 development and blocks the full pipeline from being operational.
Parent Epic: This relates to the plan lifecycle pipeline. Linking to the execution pipeline work.
Completeness: The issue has all required sections per CONTRIBUTING.md — Metadata, Background, Root Cause, Acceptance Criteria, and Definition of Done. The acceptance criteria are detailed and serve as implicit subtasks.

Next step: This issue is now verified and ready for implementation. The branch name bugfix/m1-plan-execute-sandbox-root and commit message are pre-defined in the Metadata section.

## Triage: Verified This is a well-documented, critical bug filed by @brent.edwards during development of #1249 (M1 acceptance E2E test hardening). The root cause analysis is thorough and the code path is clearly traced. **Assessment:** - **Priority**: Critical — confirmed. This bug means the core plan execute pipeline silently discards LLM-generated files, which breaks the M1 acceptance criterion ("Execute/apply runs via actor-based LLM path producing tool invocations and a ChangeSet"). - **Milestone**: Assigning to **v3.7.0**. While this is an M1-era bug, it was discovered during v3.7.0 development and blocks the full pipeline from being operational. - **Parent Epic**: This relates to the plan lifecycle pipeline. Linking to the execution pipeline work. - **Completeness**: The issue has all required sections per CONTRIBUTING.md — Metadata, Background, Root Cause, Acceptance Criteria, and Definition of Done. The acceptance criteria are detailed and serve as implicit subtasks. **Next step**: This issue is now verified and ready for implementation. The branch name `bugfix/m1-plan-execute-sandbox-root` and commit message are pre-defined in the Metadata section.

freemo added this to the v3.7.0 milestone

2026-04-02 16:49:03 +00:00

freemo added

and removed

labels

2026-04-02 16:49:14 +00:00

freemo referenced this issue

2026-04-02 16:56:18 +00:00

[Automated] Product Build Session State #1314

freemo referenced this issue

2026-04-02 16:57:45 +00:00

[Automated] Product Build Session State #1314

freemo referenced this issue

2026-04-02 16:58:47 +00:00

test(e2e): update m1_acceptance.robot #1249

freemo self-assigned this

2026-04-02 18:45:24 +00:00

freemo added the

MoSCoW

Must have

label

2026-04-02 18:59:16 +00:00

freemo commented

2026-04-02 18:59:25 +00:00

Owner

Label compliance fix applied:

Added missing label: MoSCoW/Must have
Reason: This is a Priority/Critical bug in State/Verified for the v3.7.0 milestone. Wiring sandbox_root into the plan execute pipeline is a core M1 acceptance criterion — it qualifies as Must Have.

Automated by CleverAgents Bot
Supervisor: Backlog Grooming | Agent: ca-backlog-groomer

Label compliance fix applied: - Added missing label: `MoSCoW/Must have` - Reason: This is a `Priority/Critical` bug in `State/Verified` for the v3.7.0 milestone. Wiring `sandbox_root` into the plan execute pipeline is a core M1 acceptance criterion — it qualifies as Must Have. --- **Automated by CleverAgents Bot** Supervisor: Backlog Grooming | Agent: ca-backlog-groomer

freemo referenced this issue

2026-04-02 19:00:16 +00:00

[Automated] Backlog Groomer Session State — v3.7.0 #1463

freemo modified the milestone from v3.7.0 to v3.2.0

2026-04-02 21:58:02 +00:00

freemo removed the