[P0] Executor over-engineers small tasks — creates venvs / pip-installs deps #33

New issue

Open

opened 2026-05-23 21:52:20 +00:00 by thabeta · 0 comments

thabeta commented

2026-05-23 21:52:20 +00:00

Owner

Problem
For a self-contained task ("build a small unittest library") the executor ran python3 -m venv .venv + pip install pytest + wrote pyproject.toml, dumping ~300 dependency files into the workspace and burning ~2 min. It should write a stdlib self-test and never create virtualenvs or install packages for self-contained work unless explicitly asked.

Evidence

Live job c_1779571735407_… created .venv/** (all of pytest/pip/setuptools).
Executor prompt: crates/hero_shrimp_engine/src/orchestration/autonomy/prompt_builder.rs (AutonomyMode::Execute).

Proposed fix
Add an Execute-mode rule: don't create virtualenvs or install packages for a self-contained library/script; verify with the stdlib or what's already present. (UI now hides vendored paths, but that only masks the symptom.)

Filed from a comparative audit of Hero Shrimp vs Qwen-Code / kimi-cli / picoclaw (2026-05-23). Severity in title: P0=correctness/trust, P1=reliability/UX, P2=cleanup.

**Problem** For a self-contained task ("build a small unittest library") the executor ran `python3 -m venv .venv` + `pip install pytest` + wrote `pyproject.toml`, dumping ~300 dependency files into the workspace and burning ~2 min. It should write a stdlib self-test and never create virtualenvs or install packages for self-contained work unless explicitly asked. **Evidence** - Live job `c_1779571735407_…` created `.venv/**` (all of pytest/pip/setuptools). - Executor prompt: `crates/hero_shrimp_engine/src/orchestration/autonomy/prompt_builder.rs` (AutonomyMode::Execute). **Proposed fix** Add an Execute-mode rule: don't create virtualenvs or install packages for a self-contained library/script; verify with the stdlib or what's already present. (UI now hides vendored paths, but that only masks the symptom.) --- _Filed from a comparative audit of Hero Shrimp vs Qwen-Code / kimi-cli / picoclaw (2026-05-23). Severity in title: P0=correctness/trust, P1=reliability/UX, P2=cleanup._