[rennf93]

Follow-up to trail #65 (change 213): since #1355, make codegen in a task worktree imports the clone root's code instead of the worktree's.

What happened

On the live run of 2026-11-03, task f28b1d75 (PR #1359, two doc files) went round the PM review six times:

1. Every push runs the project's codegen_command (make codegen for roboco-api) before it goes out (git.py _run_codegen_and_commit).
2. fix(workspace): keep a worktree's path out of the shared venv's shebangs #1355 runs a worktree's commands in the workspace container with UV_PROJECT_ENVIRONMENT set to readlink -f .venv, which is the clone root's venv (workspace_container.py:89,138). So sys.prefix is <clone>/.venv.
3. The import hook only acts when sys.prefix is <clone>/.worktrees/<id>/.venv (worktree_imports.py _worktree_whose_venv_is), so it left sys.path alone and the script imported the clone root's roboco.
4. fe-doc's clone root sits on slave at 3f96eba (2026-08-31). Its codegen wrote agents/prompts/_generated/verbs.md as blob c3870cb3. The PR base, slave and master all hold eca32f19.
5. fe-doc committed "regenerate generated artifacts" on every i_documented, fe-pm sent it back ("PR diff holds only CHANGELOG.md and docs/positioning/..."), and fe-dev-2 reverted. PR [f28b1d75] Write ETR positioning note and CHANGELOG entry #1359 has 6 of each.

Fix

When sys.prefix is a clone root's own .venv, the hook now points the clone root's sys.path entries at the worktree that holds the working directory. The container already runs the command there (-w is the worktree). A process outside .worktrees/<id>, or a prefix that is not that clone's .venv, is left alone as before. A vanished working directory counts as none.

Behavior change: code run in a worktree on the canonical venv imports the worktree's source. Nothing else changes. #1355 stays as it is, since reverting it brings back worktree paths in the shared venv's shebangs.

Not in this PR: why the clone roots stopped fast-forwarding to slave after Aug 31. I didn't check.

Verified

• New tests in tests/unit/services/test_workspace_worktree_imports.py: a parametrized one for the canonical prefix (6 cases), and one that runs a script in a worktree through the clone root's venv python and checks it imports the worktree's package. On the old hook, 3 of them fail (3 failed, 17 passed). With the fix: 20 passed.
• pytest tests/unit/services/ -k 'workspace or worktree': 331 passed, 4266 deselected.
• ruff check and ruff format --check clean, mypy clean on both files, xenon exits 0, vulture and bandit report nothing, verify_codebase_map.py reports 0 errors and 0 warnings, and _complete_map.md is in sync.

[github-actions]

Thanks for opening your first pull request on RoboCo!

Quick checklist before review (most of these are enforced by CI, but worth a glance):

• make quality — ruff format check, ruff check, mypy, pytest (≥80% coverage), and the rest of the gate
• Panel changes pass pnpm lint and pnpm exec tsc --noEmit (run from panel/)
• No # noqa / # type: ignore shortcuts; pre-existing violations in touched files are fixed
• Added an entry under ## [Unreleased] in CHANGELOG.md
• Signed the CLA (the bot will prompt you on this PR)
• Signed your commits — master requires verified signatures (SSH signing setup)
• Updated any affected docs under docs/

See CONTRIBUTING.md for the full workflow and the Code of Conduct for the community standards we follow.

Welcome aboard — a maintainer will review shortly.