Lap Tran
bc7daf8590
- **Restore 82 completed tasks** from tasks/complete/ back to tasks/ top level (all <7 days old per the cleanup policy; premature bulk archive was fixed). - **Dashboard: fix scroll-reset on auto-refresh** — renderBoard rebuilds the board via innerHTML every 2s, destroying each column-body's scrollTop. Now snapshots column-body scrollTop + board.scrollLeft + view.scrollTop before rebuild and restores after (matched by PHASE_GROUPS index). - **Dashboard UI additions** (pre-existing unstaged work): approval section cards, transition buttons, inline artifact editor (textarea for writing missing SPEC/VERDICT/etc from the detail modal). - **Bind ornith as Implement model** — config.md: Model explicit to omlx/Ornith-1.0-35B-4bit-mlx, context window 32768. Interactive autopilot already used ornith via opencode default; now explicit. - **Fix cleanup stub** — automaton-cleanup.sh had a stale --project arg pointing at a pytest temp dir (test isolation leak). Rewired to point at ~/.automaton. - **Fix plist-isolation test** — test asserted host plist doesn't exist, but a real install creates it. Now snapshots mtime before run, asserts unchanged after (only a write during the test counts as bleed). - **New Playwright smoke test** (tests/test_dashboard_ui.py) — 2 tests: board renders tasks, column scroll survives auto-refresh tick. Verified the test fails without the scroll fix (scrollTop resets to 0). Skipped via importorskip when playwright is absent (main CI stays green). - **Clarify SI loop scope in README** — new-project onboarding section documents the framework-scoped self-improvement loop and options (leave/pause/create project loop). - **CHANGELOG** documents all changes including the known model-divergence gap (mde tasks marked complete but per-role model binding was never implemented).
13 lines
1.3 KiB
Markdown
13 lines
1.3 KiB
Markdown
# Adversarial Bug Report: Autopilot Gate Integration
|
|
|
|
## Deep Review
|
|
The gate-check loop in orchestrate.md replaces the previous drive_all() pseudocode with an explicit phase-by-phase process. Each phase is validated before and after. Approval gates are hard stops, not soft suggestions.
|
|
|
|
## Potential Issues
|
|
1. **Self-approval risk**: In autopilot mode, the orchestrator prompt says "STOP and wait for user approval" at approval gates. However, the orchestrator is the same agent that completes the phase. A non-compliant orchestrator could skip the approval gate and call `--approve` itself. Mitigation: `--approve` is designed to require explicit user action, but the enforcement is prompt-based within a single agent session.
|
|
|
|
2. **Session context loss at approval pause**: When autopilot pauses for user approval and the user returns in a new session, the orchestrator must re-read `.state` to know where it left off. This works correctly but depends on the `.state` file being written before the pause.
|
|
|
|
3. **No timeout on approval pauses**: If the user never returns to approve a phase, the task is stuck in `:awaiting_approval` indefinitely. This is by design (user must approve), but there's no notification mechanism.
|
|
|
|
## Verdict: PASS — the self-approval risk is an inherent limitation of prompt-based enforcement, not a bug. |