mirror of
https://github.com/openswarm-ai/openswarm.git
synced 2026-09-12 12:47:42 +02:00
ciregenzandClaude Fable 5
dfe819b095
125 tasks, one external grader (WOB_REWARD_GLOBAL through BrowserGym), one recorder book, agents never grade themselves. Arms: our shipped perception ported line-for-line (deterministic + LLM), a browser-use-shaped flat-dump arm on the same model, the real browser-use agent attached over CDP to the same seeded page, and Stagehand wiring. Techniques already ingested from the competitor diffs, each landing measured points: clickable-but-unroled elements (canvas/svg -- their flat dump solved spatial tasks purely because the canvas was in it), coordinate actions (our click_point equivalent), input twins exempt from dedupe (two same-named password boxes are two fields), select options rendered on the row, and the page-text panel (our BrowserGetText equivalent). v1 25.6% -> v2 32.0% on identical tasks/model, wall clock down 3.8s -> 3.2s. Supervision is out-of-process: SIGALRM exceptions got swallowed inside stack retry loops twice, so a watcher kills the process tree when the recorder stops moving and resumes exactly the missing tasks. 502-dead LLM lanes book as infra, never as skill. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WsbS5x2rYsMDxP2kW3qqmQ
85 lines
2.1 KiB
Plaintext
85 lines
2.1 KiB
Plaintext
.DS_Store
|
|
.worktrees
|
|
.env
|
|
.env.*
|
|
!.env*.example
|
|
.local-stash/
|
|
|
|
# Playwright e2e artifacts (per-run traces, screenshots, reports, raw results).
|
|
e2e/traces/
|
|
e2e/playwright-report/
|
|
e2e/test-results/
|
|
e2e/results.json
|
|
# Playwright also drops a test-results/ at the repo root when run from here.
|
|
test-results/
|
|
backend/data/**
|
|
!backend/data/outputs/
|
|
!backend/data/outputs/*.json
|
|
|
|
# Electron build artifacts
|
|
electron/dist/
|
|
electron/python-env/
|
|
electron/build-staging/
|
|
electron/node_modules/
|
|
# electron/package-lock.json is intentionally COMMITTED (tracked): the build
|
|
# runs `npm ci`, which needs the lockfile in-repo. Do not re-add this ignore.
|
|
electron/build-info.json
|
|
|
|
# Router is fetched from npm at build time into electron/build-staging/router.
|
|
# No local router/ directory is tracked.
|
|
|
|
# Pre-installed MCP server dependencies
|
|
backend/npm-servers/*/node_modules/
|
|
|
|
# Frontend build output
|
|
frontend/dist/
|
|
# Bundled uv binaries (downloaded during build)
|
|
backend/uv-bin/
|
|
# Pre-built webapp-template node_modules archive (generated by
|
|
# scripts/build-template-archive.sh, optionally bundled into the DMG).
|
|
backend/apps/outputs/webapp_template_cache/
|
|
# Python virtualenvs anywhere (a root .venv was accidentally committed once)
|
|
.venv/
|
|
backend/.venv/
|
|
.account-factory
|
|
openswarm-cloud
|
|
.openswarm-cloud
|
|
# Top-level only — Haik's PostHog dashboard webapp clone. Anchored with
|
|
# leading slash so this doesn't accidentally ignore backend/apps/analytics/.
|
|
/analytics
|
|
.claude/
|
|
# Personal/per-machine AI working notes. Shared, enforced conventions live in committed
|
|
# CLAUDE.md files (root + per-package); personal notes go in CLAUDE.local.md instead.
|
|
CLAUDE.local.md
|
|
/CLAUDE.md
|
|
# Local-only operator helpers (never commit)
|
|
scripts/set-fly-*.sh
|
|
# Local-only background dev-team state/docs (personal, never commit)
|
|
docs/ops/
|
|
|
|
# Python bytecode (regenerates on every import)
|
|
__pycache__/
|
|
*.pyc
|
|
*.pyo
|
|
|
|
# Test/lint caches (all auto-rebuild on next run)
|
|
.pytest_cache/
|
|
.mypy_cache/
|
|
.ruff_cache/
|
|
.coverage
|
|
htmlcov/
|
|
|
|
# Editor noise — per-developer, never useful in repo
|
|
.idea/
|
|
.vscode/
|
|
*.swp
|
|
*~
|
|
.envrc
|
|
|
|
# OS detritus
|
|
Thumbs.db
|
|
ehthumbs.db
|
|
desktop.ini
|
|
frontend/tsconfig.tsbuildinfo
|
|
e2e/browser-v3/arena/data/
|