Last updated: 2026-06-14
- Preserved V1 open-source evaluation baseline
- Identified strong_multi_goal task family as V2 focus
- Established evaluation infrastructure (agent_eval_session, repeated_run_summary)
- Ambiguity-aware retrieval (navigation_prune mode)
- Structured subgoal decomposition and shallow probing
- Shared-workspace clustering (path-prefix + relation-overlap)
- Primary/backup branch selection with micro-context retention
- Default-output trimming, stable-entrypoint anchoring
- Repeated-run evaluation: 3 strong_multi_goal tasks, all net-positive
- Status: Frozen — commit
ac67177
- Parser tests: +111 tests (Python AST + JS/TS regex)
- Store tests: +41 tests (SQLite engine, FTS, schema migrations)
_now_utcdedup: 9 copies → sharedcodewell.utils.now_utc- N+1 INSERT batching:
executemanyfor symbols/imports/call_edges - Exception handling audit + CHANGELOG stale claims fix
- Scanner/commands/repair_audit tests: +40 tests
- Dedupe unification + 10 core function docstrings
- Tests: 295 → 487 (+65%)
- Status: Complete — commit
ac67177
- Security/config, validation/schema, frontend auth, ORM/database, framework config
- 3×3 repeated-run protocol: 30 runs total
- Initial results: 3/5 elapsed-positive, 5/5 tools-positive
- Post-fix (security + pipe keyword expansion): 5/5 all-positive
- Status: Complete — commits
9de45bd,5e61f29
- Merged feature branch to main with --no-ff
- Tagged
v2.0.0-alpha - Status: Complete — commit
f61dc3c
| # | Task | Domain | Goal Shape |
|---|---|---|---|
| 1 | nestjs_file_upload_type_validation_wiring | File handling | weak_multi_goal |
| 2 | nestjs_mail_i18n_context_null_safety_wiring | Email/notifications | strong_multi_goal |
| 3 | nestjs_serializer_interceptor_global_wiring | Data serialization | weak_multi_goal |
- 3 tasks × 3 repeats × 2 conditions = 18 runs
- Expected duration: ~45-90 minutes
- No regression in existing 11 tasks
- At least 2/3 new tasks elapsed-positive
- Tools-positive on all 3 new tasks
Why: V2 claims Python support via python_parser.py and SOURCE_LANGUAGE_BY_SUFFIX,
but all 14 evaluation tasks are TypeScript/JavaScript. Python AST parsing is fundamentally
different from JS/TS regex parsing, and the retrieval pipeline should be validated on Python.
Available repos:
eval-repos/pypi/click/— CLI framework (commands, options, arguments, types)eval-repos/pypi/requests/— HTTP library (sessions, adapters, auth, redirects)eval-repos/pypi/rich/— Terminal UI (console, tables, progress, layout)
Approach:
- Study each library's structure (2-3 hours)
- Identify realistic multi-goal bugs (1-2 per library)
- Create task definitions and goal labels
- Run 3×3 repeated-run protocol
Target: 2-3 Python tasks across 2 libraries
Success Criteria:
- At least 2/3 tasks elapsed-positive
- Comparable tool/token efficiency to TS tasks
- No regression in existing tasks
Effort estimate: 0.5-1 day
Items:
- Clean up
freeze-snapshots/directory - Add
.pytest_tmp/to.gitignore - Verify CI/CD pipeline (GitHub Actions)
- Run pre-release check script
- Package smoke test
- Final documentation pass
- Tag
v2.0.0-beta
Effort estimate: 0.5 day
| Domain | Tasks | Status |
|---|---|---|
| Middleware/Auth | 2 | ✅ V2 baseline |
| Route/Guard | 1 | ✅ V2 baseline |
| Security/Config | 1 | ✅ Phase 3 |
| Validation/Schema | 1 | ✅ Phase 3 |
| Frontend Auth | 1 | ✅ Phase 3 |
| ORM/Database | 1 | ✅ Phase 3 |
| Framework Config | 1 | ✅ Phase 3 |
| File Handling | 1 | 🔄 Phase 4 |
| Email/Notifications | 1 | 🔄 Phase 4 |
| Data Serialization | 1 | 🔄 Phase 4 |
| CLI/Commands | 0 | ⏳ Phase 5 (Python) |
| HTTP Client | 0 | ⏳ Phase 5 (Python) |
| Terminal UI | 0 | ⏳ Phase 5 (Python) |
| Language | Tasks | Status |
|---|---|---|
| TypeScript/JavaScript | 11 | ✅ Complete |
| Python | 0 | ⏳ Phase 5 |
| Framework | Tasks |
|---|---|
| NestJS | 6 |
| Express | 2 |
| AdonisJS | 2 |
| React/Next.js | 1 |
| Click (Python) | 0 |
| Requests (Python) | 0 |
| Metric | Value |
|---|---|
| Total tests | 487 |
| Passing | 484 |
| Pre-existing failures | 3 (test_self_evaluation.py) |
| New tests since V2 start | +192 (+65%) |
9490225 feat(phase4): add file upload, mail i18n, and serializer interceptor tasks
f61dc3c merge: V2 multi-goal retrieval baseline (main)
5e61f29 fix(v2): add security and pipe keyword sets
9de45bd feat(phase3): add benchmark expansion tasks and evaluation infrastructure
ac67177 feat(v2): freeze multi-goal retrieval baseline with quality hardening
v2.0.0-alpha ← tag
Now: Phase 4 evaluation (3 tasks × 3 repeats)
Update V2_PROGRESS doc
Next: Phase 5 Python tasks (research + create + evaluate)
Then: Phase 6 Production readiness
Tag v2.0.0-beta
Future: Optional LLM planner (when rule expansion returns diminish)
Optional embedding clustering (when path clustering fails)