brindle 0.5.2__tar.gz → 0.5.4__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {brindle-0.5.2 → brindle-0.5.4}/PKG-INFO +34 -36
- {brindle-0.5.2 → brindle-0.5.4}/README.md +33 -35
- {brindle-0.5.2 → brindle-0.5.4}/pyproject.toml +1 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/agents.py +50 -29
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/antigravity.py +1 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/autopilot.py +13 -18
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/check_scope.py +109 -11
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cli.py +22 -35
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cloud_login.py +21 -8
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/config.py +21 -8
- brindle-0.5.4/src/brindle/connections.py +533 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/api-schema.json +1912 -802
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/lifecycle.py +2 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/routes_actions.py +256 -95
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/routes_read.py +113 -22
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/session_review.py +25 -25
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/db.py +35 -3
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/gates.py +31 -14
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/key_check.py +27 -10
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/keys_cmds.py +61 -3
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/keystore.py +82 -3
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/runner.py +2 -2
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/org_permissions.py +22 -2
- brindle-0.5.4/src/brindle/permission_grants.py +1523 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/permission_groups.py +18 -15
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/permissions.py +207 -69
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/team_policy.py +55 -3
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/profiles.py +7 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/providers.py +23 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/providers_install.py +6 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/dto.py +115 -1
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/inbox.py +41 -3
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/permissions.py +47 -4
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/sessions.py +3 -7
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/sessions.py +6 -5
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/tmux.py +50 -2
- brindle-0.5.2/src/brindle/permission_grants.py +0 -803
- {brindle-0.5.2 → brindle-0.5.4}/.gitignore +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/LICENSE +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/SCHEDULE-A +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/__main__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/account.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/airgap.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/assembler.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/assets/braid_dark.png +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/assets/braid_light.png +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/budget.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/backend-auditor.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/coordinator.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-codex.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-fable.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-gemini.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-heavy.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-light.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-local.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-opencode.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer-sonnet.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/developer.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/qa.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/reviewer-codex.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/reviewer-gemini.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/reviewer-local.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/reviewer-opencode.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/reviewer.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/subagent.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_agents/supervisor.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_rules/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_rules/security/backend.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_rules/style/minimal-diff.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/builtin_rules/tests/only.md +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/bus.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/ci_adapters.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/ci_client.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/ci_federation.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/ci_hosts.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/clone.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cloud_credential.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/codemap.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/codemap_docs.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/codex_hook.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/company_identity.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/company_login.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/conflicts.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/coordinator.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cost.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cost_estimate.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/cull.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/app.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/auth.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/events.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/holds.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/peer.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/routes_org_permissions.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/schema.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/daemon/terminal.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/debuglog.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/demo.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/detect.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/doctor.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/entry.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/escapes.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/events.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/git.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/guardrails.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/history.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/identity_lock.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/inbox.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/kb.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/knowledge.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/launch_health.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/learned_rules.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/learning.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/legacy.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/local_setup.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/login_pause.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/mcp_server.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/model_access.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/model_availability.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/catalog.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/client.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/loop.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/lsp.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/permissions.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/serve.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/native/tools.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/opencode.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pane_auth.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/permission_asks.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/permission_routing.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/picker.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pipeline.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/plugins.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/policy.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pool.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/preflight.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pricing.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/_files.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/account.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/audit_chain.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/audit_export.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/auth.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/brief_learning.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/cost_centers.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/credentials.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/features.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/keys.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/learning.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/license.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/loopback.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/managed_models.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/org_profiles.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/orgkey.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/rollout.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/route_health.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/settings_sync.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/status.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/pro/team_events.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/procs.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/prompt_answer.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/provider_homes.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/quota.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/repo_cmds.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/repos.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/rewind.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/rule_checks.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/savings.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/scratch.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/secrets.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/self_serve.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/errors.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/service/routing.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/services.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/sidebar_follow.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/sidebar_logo.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/signin_pause.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/status_cache.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/tasks.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/updates.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/upgrade.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/usage.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/view.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/wake.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/watch.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/watch_costs.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/__init__.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/chat.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/dashboard.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/diff.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/server.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/static/app.js +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/web/static/index.html +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/worker_models.py +0 -0
- {brindle-0.5.2 → brindle-0.5.4}/src/brindle/workspaces.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: brindle
|
|
3
|
-
Version: 0.5.
|
|
3
|
+
Version: 0.5.4
|
|
4
4
|
Summary: Run coding agents in parallel, each on its own git branch, with a supervisor that delegates, reviews, and merges
|
|
5
5
|
Project-URL: Homepage, https://pawdelta.com/brindle/
|
|
6
6
|
Project-URL: Documentation, https://pawdelta.com/brindle/docs/recommended-use
|
|
@@ -70,7 +70,7 @@ service ([details](#license)).
|
|
|
70
70
|
- **Delegation built in.** Agents get a `brindle` MCP server: `assign` work to
|
|
71
71
|
parallel workers with the files they'll touch and what they depend on, `handoff`
|
|
72
72
|
and wait, `send_message`, then review and merge without spending the supervisor's turns.
|
|
73
|
-
- **Walk away.**
|
|
73
|
+
- **Walk away.** brindle keeps going until every milestone is verified, pauses
|
|
74
74
|
at your usage limit and resumes after the reset, and asks you only when it
|
|
75
75
|
needs a decision. Quit and `brindle continue` later; workers resume too.
|
|
76
76
|
- **Any model, including local ones.** Claude Code, Codex, Google Antigravity, or
|
|
@@ -158,8 +158,7 @@ switch windows or sessions; you can still move it manually or use `brindle attac
|
|
|
158
158
|
to choose a brindle window.
|
|
159
159
|
|
|
160
160
|
**Sidebar keys.** `?` shows them in the sidebar too. ◆ marks an agent that needs
|
|
161
|
-
you: stuck on a prompt
|
|
162
|
-
worker whose branch is waiting for your review. Under autopilot, a reported worker shows
|
|
161
|
+
you: stuck on a prompt or a supervisor with a question. A reported worker shows
|
|
163
162
|
a dim ◇ instead, because the supervisor reviews it.
|
|
164
163
|
|
|
165
164
|
**Copying chat text.** Drag with the mouse in the chat: the selection stays inside
|
|
@@ -263,7 +262,7 @@ things itself instead of starting workers.
|
|
|
263
262
|
2. **Give it a goal with checks.** For anything bigger than one sitting, tell the
|
|
264
263
|
supervisor what "done" means in commands (a test file, `npm test`, a build), or
|
|
265
264
|
write `.brindle/goals.md` yourself. Milestones with real checks are what let
|
|
266
|
-
|
|
265
|
+
brindle keep going without you and stop when the work is actually done.
|
|
267
266
|
3. **Set `checks` in `.brindle/config.json`** (usually your full test suite) so no
|
|
268
267
|
branch merges red, and a `setup` if new worktrees need `npm install` or similar.
|
|
269
268
|
4. **Watch the sidebar, not every window.** It flags agents waiting on you (◆);
|
|
@@ -281,9 +280,9 @@ task to one area of the code with its own test command, and let the supervisor r
|
|
|
281
280
|
to `max_agents` workers at once. Use a cheaper profile (see [Cheap workers](#cheap-workers))
|
|
282
281
|
for mechanical edits and a Codex reviewer for a second model's opinion.
|
|
283
282
|
|
|
284
|
-
##
|
|
283
|
+
## Driving toward a verified goal
|
|
285
284
|
|
|
286
|
-
`brindle`
|
|
285
|
+
`brindle` drives the supervisor toward a verified goal. Tell it what we're building,
|
|
287
286
|
or put the goal in `.brindle/goals.md`, and it works like a project manager:
|
|
288
287
|
|
|
289
288
|
1. **Goal and milestones.** The goal is split into milestones, and each one has
|
|
@@ -309,7 +308,7 @@ or put the goal in `.brindle/goals.md`, and it works like a project manager:
|
|
|
309
308
|
approves it, or (`approved=false`, with feedback) sends it back for a
|
|
310
309
|
revision. For Claude workers brindle's PreToolUse hook refuses Edit, Write
|
|
311
310
|
and NotebookEdit until the plan is approved; other CLIs are only told to
|
|
312
|
-
wait.
|
|
311
|
+
wait. brindle doesn't count a worker waiting on approval as stalled, and
|
|
313
312
|
reminds the supervisor about plans awaiting a decision.
|
|
314
313
|
3. **Gated merges.** A branch merges only when everything is committed, a
|
|
315
314
|
reviewer agent has approved that exact commit, your pre-commit hooks pass,
|
|
@@ -353,7 +352,7 @@ for byte as you wrote it. Since `goals.md` may be committed, it stays free of
|
|
|
353
352
|
session data: no agent or session ids, check output, notes, usage or
|
|
354
353
|
questions ever go in. A status line is information only: a new session
|
|
355
354
|
starts every milestone as pending and re-runs the checks, never trusting it.
|
|
356
|
-
A session stops writing when it's handed over
|
|
355
|
+
A session stops writing when it's handed over or paused,
|
|
357
356
|
or once its goal was replaced from the chat. From a linked worktree, it writes
|
|
358
357
|
the `goals.md` it loaded (the main checkout's), never another session's. A
|
|
359
358
|
write that fails never fails the check.
|
|
@@ -363,15 +362,14 @@ A new session takes a hand-written `goals.md` over by moving it to its own
|
|
|
363
362
|
its deletion is left for you to commit or restore; brindle doesn't count it as
|
|
364
363
|
an uncommitted change, so it never blocks a merge or a milestone's check.
|
|
365
364
|
|
|
366
|
-
`brindle
|
|
367
|
-
|
|
368
|
-
|
|
369
|
-
starts without it.
|
|
365
|
+
`brindle goal` shows progress and `brindle goal check` runs the checks now.
|
|
366
|
+
Pause the session to stop it. The deprecated `--autopilot`/`--no-autopilot`
|
|
367
|
+
flags and the `autopilot` config key are ignored: goal driving is always on.
|
|
370
368
|
|
|
371
|
-
When your Claude usage reaches `usage_limit`,
|
|
369
|
+
When your Claude usage reaches `usage_limit`, brindle pauses for usage: its
|
|
372
370
|
running Claude workers stop (worktrees, branches, queued messages and sessions
|
|
373
371
|
are kept) and the sidebar says "paused for usage until <time>". Once the usage
|
|
374
|
-
window resets, brindle restarts those workers on its own,
|
|
372
|
+
window resets, brindle restarts those workers on its own, resumes goal driving
|
|
375
373
|
again and tells the supervisor what it resumed. Speaking to the supervisor
|
|
376
374
|
doesn't end the pause.
|
|
377
375
|
|
|
@@ -391,9 +389,9 @@ your own status line prints, so what you see doesn't change.
|
|
|
391
389
|
| `brindle new BRANCH [-b BASE] [-p PROMPT] [--no-agent]` | worktree + branch + a supervisor in it |
|
|
392
390
|
| `brindle continue [ID]` / `brindle -c` | resume a paused session (default: the most recent) |
|
|
393
391
|
| `brindle sessions` / `brindle prune` | list paused sessions / apply retention rules, remove merged or missing worktrees, close stale stopped agents whose work is merged or gone, and remove leftover tmux sessions |
|
|
394
|
-
| `brindle start [-p PROMPT] [--no-watch] [
|
|
392
|
+
| `brindle start [-p PROMPT] [--no-watch] [-b BRANCH] [-w PATH]` | the same, with options; `-b`/`-w` run it in that branch's worktree (created if needed, or the one you already made), which gets the repo's `.brindle` config |
|
|
395
393
|
| `brindle handover --to BRANCH\|PATH [-n NOTE]` | hand the session to a new supervisor there: goal and milestones, workers, queued tasks and your note move across; the old one is paused |
|
|
396
|
-
| `brindle
|
|
394
|
+
| `brindle goal [check]` | the goal's progress; run the checks now |
|
|
397
395
|
| `brindle audit-decision FINDING [-r RATIONALE]` | record acceptance of an audit finding with rationale so subsequent audits respect it |
|
|
398
396
|
| `brindle delegation [conservative\|balanced\|fast]` | how readily the supervisor delegates: fewest tokens, the default, or quickest |
|
|
399
397
|
| `brindle transfer [REPO] [--from SESSION] [-b BRANCH]` | move a scratch session's work into a real repo |
|
|
@@ -443,7 +441,7 @@ With no `WS` argument, commands act on the workspace you're in.
|
|
|
443
441
|
`brindle web` serves the current session to your browser: workspaces, agents
|
|
444
442
|
and their status, and the goal's milestones with each one's last check. Run
|
|
445
443
|
it from a terminal in the session's tmux session to show that session's
|
|
446
|
-
goal; elsewhere it shows the newest
|
|
444
|
+
goal; elsewhere it shows the newest session
|
|
447
445
|
in the repo. The page refreshes itself every couple of seconds.
|
|
448
446
|
|
|
449
447
|
It listens on `127.0.0.1` only, on a free port (or `--port N`), and prints a
|
|
@@ -475,7 +473,7 @@ knowing them helps when you tell the supervisor how to work.
|
|
|
475
473
|
| `approve_plan` | supervisor | approve a worker's plan, or send it back with feedback (`approved=false`) |
|
|
476
474
|
| `answer_permission` | supervisor | allow, deny or escalate (with a reason) a worker's permission ask before the person sees it |
|
|
477
475
|
| `complete_subagent` | supervisor | record the result of a `subagent`-profile task |
|
|
478
|
-
| `set_goal` / `get_progress` / `check_milestone` | supervisor |
|
|
476
|
+
| `set_goal` / `get_progress` / `check_milestone` | supervisor | the session's goal, its progress, and running the checks |
|
|
479
477
|
| `record_audit_decision` | supervisor | record acceptance of an audit finding with rationale so subsequent audits respect it |
|
|
480
478
|
| `who_is_working` | supervisor | other sessions in this repo: their goals and running tasks, optionally only those matching `paths` or a `query` |
|
|
481
479
|
| `record_decision` / `decisions` | supervisor (`decisions` also workers) | record a durable decision about the system / list the accepted ones that apply |
|
|
@@ -544,7 +542,7 @@ Set in `~/.brindle/config.json`, `.brindle/config.json` or `config.local.json`
|
|
|
544
542
|
take the task, its goal or the month over a limit (a task's cost is estimated
|
|
545
543
|
from this repo's finished tasks on that profile, else a typical token mix at
|
|
546
544
|
list prices), brindle skips that candidate and uses the next cheaper one. If
|
|
547
|
-
none fits it refuses the assign, says why and
|
|
545
|
+
none fits it refuses the assign, says why and asks you. A
|
|
548
546
|
profile you name yourself is never second-guessed. Running workers are
|
|
549
547
|
reported to the supervisor at 80% of `task_usd` and again at 100%; set
|
|
550
548
|
`"stop": true` to close a worker at 100% instead of only warning. Without the
|
|
@@ -633,7 +631,7 @@ brindle coordinator stop # close it and turn it off for this che
|
|
|
633
631
|
brindle keeps one place (`~/.brindle/quota.json`) that knows how close each
|
|
634
632
|
provider is to its subscription limit, and shows a note per provider that has
|
|
635
633
|
data (e.g. `Codex at 82% of its weekly limit, resets Thu 9:00am`) in `assign`
|
|
636
|
-
replies, `get_progress`, the sidebar's
|
|
634
|
+
replies, `get_progress`, the sidebar's goal block and `brindle doctor`.
|
|
637
635
|
It only uses what the CLIs write locally, and never reads a CLI's login
|
|
638
636
|
token or auth files or calls a provider's servers:
|
|
639
637
|
|
|
@@ -666,16 +664,16 @@ A linked git worktree you made yourself doesn't have the git-ignored parts of
|
|
|
666
664
|
}
|
|
667
665
|
```
|
|
668
666
|
|
|
669
|
-
|
|
667
|
+
Goal driving, merge gates and cleanup:
|
|
670
668
|
|
|
671
669
|
| Key | Default | |
|
|
672
670
|
|---|---|---|
|
|
673
|
-
| `autopilot` | `true` | start the supervisor with autopilot on |
|
|
674
671
|
| `claude_permission_mode` | `"default"` | the `--permission-mode` for Claude Code sessions whose profile sets no `permission_mode`; the default holds permission requests for the brindle app, set `"auto"` to opt into auto mode. Only read from `~/.brindle/config.json` |
|
|
675
672
|
| `auto_update` | `false` | run `uv tool upgrade brindle` in the background at session start when PyPI has a newer version; restart to use it. Can also be set in `~/.brindle/config.json` for every repo |
|
|
676
673
|
| `scoped_checks` | unset (acts as `"off"`) | `"infer"` narrows a bare pytest check (`pytest ...`, `python -m pytest ...`, `uv run pytest ...`, options only) on a worker's branch to the tests `graphify affected` reaches plus the touched test files, when the repo has a stored graph; it runs in full when the mapping fails or `conftest.py`, `pyproject.toml`, `setup.cfg`, `pytest.ini`, `tox.ini` or a lockfile changed. A pytest command that already names test paths (`pytest tests/`) and every other command always run in full. `"off"` narrows nothing. It becomes the default next release; until you set it, brindle tells you once |
|
|
674
|
+
| `scoped_checks_max_files` | `40` | with `scoped_checks`, a worker's branch whose selected tests exceed this many files runs only its own changed test files and each changed module's namesake tests (`tests/test_<module>.py`, `tests/test_<module>_*.py`) instead, and the check output carries a one-line note saying so. The session branch and the full run before a PR (`gates.pr_checked`) always run everything. `0`: no cap |
|
|
677
675
|
| `checks` | `[]` | commands (plain strings) that must pass in a worker's branch before it merges. On a worker's branch every check also gets `BRINDLE_CHANGED_FILES` (the files changed since the branch left its base, one repo-relative path per line) so a script can narrow itself; the supervisor's checkout always runs every check in full (see `scoped_checks` and `brindle/check_scope.py`) |
|
|
678
|
-
| `review` |
|
|
676
|
+
| `review` | required for sessions | require a reviewer's approval before merging |
|
|
679
677
|
| `reviewer` | `"reviewer"` | the agent profile that reviews (a Codex profile gives a second model's view) |
|
|
680
678
|
| `review_profile` | none | force `request_review`'s profile, skipping its automatic cross-model pick (see below) |
|
|
681
679
|
| `pre_commit` | `true` | run [pre-commit](https://pre-commit.com) over the branch, if the repo uses it |
|
|
@@ -683,7 +681,7 @@ Autopilot, merge gates and cleanup:
|
|
|
683
681
|
| `check_timeout` | `900` | seconds each check may take |
|
|
684
682
|
| `check_concurrency` | half the logical cores, at least 2, at most 6 and 2 GB of RAM per run | check runs at once on this machine, across branches; the rest wait their turn (`0`: no cap). A run far past its last duration is reported to the supervisor |
|
|
685
683
|
| `fable_escalation` | on, unless the worker runs on a cloud account (Bedrock, Vertex or Foundry) | retry a failed heavy task once on `developer-heavy` before handing it to you; `true` turns it on for a cloud account, `false` off |
|
|
686
|
-
| `usage_limit` | `90` |
|
|
684
|
+
| `usage_limit` | `90` | brindle stops pushing on at this % of your Claude usage limit |
|
|
687
685
|
| `limit_cooldown_minutes` | `300` for Antigravity | how long a provider that hit its limit counts as unavailable |
|
|
688
686
|
| `graphify` | if the graph is there | point agents at the repo's [graphify](https://github.com/safishamsi/graphify) code map (`false` turns it off) |
|
|
689
687
|
| `graphify_auto_refresh` | `true` | automatically rebuild untracked code maps; `false` disables rebuilds while keeping queries available |
|
|
@@ -815,7 +813,7 @@ probe here; no Claude CLI or synthetic generation request is used for these chec
|
|
|
815
813
|
The real first worker task is sent once. A detached helper inspects its current
|
|
816
814
|
pane five seconds after delivery, so readiness delays do not use up that window.
|
|
817
815
|
Fresh interactive Antigravity chats without a task still receive their profile,
|
|
818
|
-
tool instructions and
|
|
816
|
+
tool instructions and goal context once after readiness; worker
|
|
819
817
|
instructions accompany the actual task in the same delivery.
|
|
820
818
|
For argv tasks, the timer starts at prompt submission or observed CLI readiness,
|
|
821
819
|
after trust/startup handling; readiness timeout does not establish delivery or
|
|
@@ -1531,7 +1529,7 @@ never handed to a worker: the supervisor gets them as "needs you".
|
|
|
1531
1529
|
|
|
1532
1530
|
**The completion audit.** A milestone's check can pass without the milestone being
|
|
1533
1531
|
done: the test asserts too little, or checks the wrong thing. So when every milestone
|
|
1534
|
-
passes,
|
|
1532
|
+
passes, brindle doesn't call the goal reached yet. A read-only reviewer audits everything since the goal was set
|
|
1535
1533
|
against the goal and each milestone, looking for what's missing, stubbed or weakly
|
|
1536
1534
|
tested. Only its approval of that commit finishes the goal. If it finds gaps, the
|
|
1537
1535
|
supervisor gets them as a message and keeps working until a fresh audit approves.
|
|
@@ -1582,7 +1580,7 @@ commands around the team's list. `~/.brindle/config.json` holds your own default
|
|
|
1582
1580
|
for every repo (any key except the command lists, like `delegation` or
|
|
1583
1581
|
`sidebar`); a repo's two files override it. With brindle Pro or higher, those
|
|
1584
1582
|
preferences (`delegation`, `sidebar`, `message_delivery`, `pr_footer`,
|
|
1585
|
-
`delete_merged_branches`, `max_agents`, `
|
|
1583
|
+
`delete_merged_branches`, `max_agents`, `plan_first`, `stale_after`,
|
|
1586
1584
|
`usage_limit`, `review_rounds`) sync across your machines (last change wins; nothing
|
|
1587
1585
|
else in the file leaves the machine); `brindle account sync` syncs now.
|
|
1588
1586
|
|
|
@@ -1690,16 +1688,16 @@ Your org's `deny_personal_keys` (Enterprise) still applies on top of `auth`.
|
|
|
1690
1688
|
|
|
1691
1689
|
#### Running unattended on API keys
|
|
1692
1690
|
|
|
1693
|
-
When running long, unattended
|
|
1691
|
+
When running long, unattended sessions with API keys:
|
|
1694
1692
|
|
|
1695
1693
|
- **Quota, token and cost measurements per provider**:
|
|
1696
|
-
-
|
|
1694
|
+
- brindle's automatic usage pause (`usage_limit: 90`) relies on the percentage numbers emitted by Claude Code's status line under subscription plans. Under API-key billing (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `GEMINI_API_KEY`), the CLIs report no subscription percentage, so subscription usage pause thresholds do not trigger; manage spend via provider account budget limits or brindle budget policies.
|
|
1697
1695
|
- Token and cost tracking: Token counts and cost estimates are captured from Claude and Codex transcripts. Antigravity (`agy`) currently has **no token/cost measurements**, so budget policies and cost tracking cannot monitor Antigravity spend; enforce usage caps directly in Google AI Studio or your Google Cloud project billing settings.
|
|
1698
1696
|
- **Runtime auth failure handling and provider differences**:
|
|
1699
|
-
- **Claude and Antigravity**: Runtime authentication failures (such as 401/403, revoked API keys, or credit/billing errors) are intercepted via StopFailure and exit hooks. Brindle marks the worker as failed, alerts the supervisor, invalidates cached sign-in checks, and pauses
|
|
1700
|
-
- **Codex (unsupported runtime auth detection)**: Codex CLI turn completion (`agent-turn-complete`) does not classify auth errors at runtime. If an OpenAI API key expires or is rejected, Codex workers will not automatically transition to `failed` or trigger
|
|
1697
|
+
- **Claude and Antigravity**: Runtime authentication failures (such as 401/403, revoked API keys, or credit/billing errors) are intercepted via StopFailure and exit hooks. Brindle marks the worker as failed, alerts the supervisor, invalidates cached sign-in checks, and pauses goal driving with a `Credentials need attention` notice rather than cycling idle nudges.
|
|
1698
|
+
- **Codex (unsupported runtime auth detection)**: Codex CLI turn completion (`agent-turn-complete`) does not classify auth errors at runtime. If an OpenAI API key expires or is rejected, Codex workers will not automatically transition to `failed` or trigger a credential pause. When running Codex workers unattended, monitor their progress via `brindle watch` or check tmux panes; if an auth failure occurs, manually restart or rotate keys.
|
|
1701
1699
|
- **Supervisor credential refresh and key rotation**:
|
|
1702
|
-
- **Recovering failed/blocked workers**: When an agent fails on an expired key or
|
|
1700
|
+
- **Recovering failed/blocked workers**: When an agent fails on an expired key or brindle pauses with a credential error, export the fresh key (or update `brindle keys set`), then run `brindle continue`. This unpauses the session and relaunches paused/failed agents with the new credentials. The replacement credentials are automatically synced to the session credential store, dynamically refreshed by the supervisor's launcher on all subsequent worker delegations (`assign`/`handoff`).
|
|
1703
1701
|
- **Proactive rotation in live sessions**: For Claude keys rotated outside brindle, set Claude Code's `apiKeyHelper` in `~/.claude/settings.json` or the project's `.claude/settings.json`. Claude Code re-runs the helper every 5 minutes (configurable with `CLAUDE_CODE_API_KEY_HELPER_TTL_MS`) and after a 401, so running workers pick up the new key. Don't also export `ANTHROPIC_API_KEY` or set a Claude key with `brindle keys set`, since an environment key takes precedence over the helper. Codex and Antigravity keys still need workers stopped, then `brindle keys set`, then `brindle continue`. Exporting a new key while an active session is healthy does not restart running workers or the supervisor (running `brindle continue` on an active healthy session simply reattaches to tmux without restarting workers). To proactively rotate credentials for the entire session and all running workers: run `brindle stop` -> update credentials in your environment or via `brindle keys set` -> run `brindle continue`.
|
|
1704
1702
|
- **Precedence**: Newly stored keys (`brindle keys set`) or session credentials updated during recovery take effect dynamically for all new panes, overriding stale launcher environments. For profile-level overrides, an explicit profile key (`env.NAME`) takes precedence over default provider credentials.
|
|
1705
1703
|
- **Antigravity permissions prerequisite**: Antigravity (`agy`) does not support approving shell commands solely via external hooks. brindle therefore runs agy workers with `--dangerously-skip-permissions` and its own hook as the gate, applying their profile's `allowed_tools` and denying the rest (see Permissions). Under an explicit `agy_approvals: "prompt"` they instead stall on approval prompts for anything `settings.json` doesn't allow, their profile's `allowed_tools` included.
|
|
@@ -2148,7 +2146,7 @@ To run workers on your own Bedrock, Vertex, Codex or Antigravity account, see
|
|
|
2148
2146
|
Codex agents report status through Codex's `notify` hook (brindle passes
|
|
2149
2147
|
`-c notify=[...]` at launch, leaving your own Codex config alone): a completed turn
|
|
2150
2148
|
marks the agent idle and delivers any queued message. Codex has no Stop hook, so for
|
|
2151
|
-
|
|
2149
|
+
a supervisor on Codex that turn end is also where brindle tells it to keep
|
|
2152
2150
|
going, with the same limit on reminders as Claude Code.
|
|
2153
2151
|
|
|
2154
2152
|
### Running workers on your own cloud
|
|
@@ -2367,7 +2365,7 @@ model's review) and the supervisor can hand it tasks.
|
|
|
2367
2365
|
|
|
2368
2366
|
`agy` has no command-line options for hooks, MCP servers or instructions, so brindle
|
|
2369
2367
|
adds three files to the checkout's `.agents/` folder: `mcp_config.json` (brindle's tools),
|
|
2370
|
-
`hooks.json` (status, messages,
|
|
2368
|
+
`hooks.json` (status, messages, goal driving) and `rules/brindle.md`. They're listed in
|
|
2371
2369
|
`.git/info/exclude`, so they never show up in `git status`. brindle adds to these files if
|
|
2372
2370
|
you already have them, and won't change one that's committed. Each agent's first
|
|
2373
2371
|
message is a short warm-up with its instructions, because `agy` connects MCP servers
|
|
@@ -2410,7 +2408,7 @@ add rules to that file yourself, for example:
|
|
|
2410
2408
|
- **Messages wake a supervisor through brindle's own hook.** Your interactive
|
|
2411
2409
|
Claude Code chat runs a small background waiter (an `asyncRewake` hook started
|
|
2412
2410
|
at `SessionStart` and at every `Stop`). When a worker reports, or brindle has a
|
|
2413
|
-
"new messages" notice or
|
|
2411
|
+
"new messages" notice or a goal nudge, the waiter hands it over and Claude
|
|
2414
2412
|
Code wakes the model with it: an idle chat starts a new turn, a busy one gets it
|
|
2415
2413
|
between steps, and the chat shows one "Stop hook feedback" line. Each message
|
|
2416
2414
|
is wrapped as `<brindle-msg nonce="..." from="developer agent w1">...</brindle-msg>`
|
|
@@ -38,7 +38,7 @@ service ([details](#license)).
|
|
|
38
38
|
- **Delegation built in.** Agents get a `brindle` MCP server: `assign` work to
|
|
39
39
|
parallel workers with the files they'll touch and what they depend on, `handoff`
|
|
40
40
|
and wait, `send_message`, then review and merge without spending the supervisor's turns.
|
|
41
|
-
- **Walk away.**
|
|
41
|
+
- **Walk away.** brindle keeps going until every milestone is verified, pauses
|
|
42
42
|
at your usage limit and resumes after the reset, and asks you only when it
|
|
43
43
|
needs a decision. Quit and `brindle continue` later; workers resume too.
|
|
44
44
|
- **Any model, including local ones.** Claude Code, Codex, Google Antigravity, or
|
|
@@ -126,8 +126,7 @@ switch windows or sessions; you can still move it manually or use `brindle attac
|
|
|
126
126
|
to choose a brindle window.
|
|
127
127
|
|
|
128
128
|
**Sidebar keys.** `?` shows them in the sidebar too. ◆ marks an agent that needs
|
|
129
|
-
you: stuck on a prompt
|
|
130
|
-
worker whose branch is waiting for your review. Under autopilot, a reported worker shows
|
|
129
|
+
you: stuck on a prompt or a supervisor with a question. A reported worker shows
|
|
131
130
|
a dim ◇ instead, because the supervisor reviews it.
|
|
132
131
|
|
|
133
132
|
**Copying chat text.** Drag with the mouse in the chat: the selection stays inside
|
|
@@ -231,7 +230,7 @@ things itself instead of starting workers.
|
|
|
231
230
|
2. **Give it a goal with checks.** For anything bigger than one sitting, tell the
|
|
232
231
|
supervisor what "done" means in commands (a test file, `npm test`, a build), or
|
|
233
232
|
write `.brindle/goals.md` yourself. Milestones with real checks are what let
|
|
234
|
-
|
|
233
|
+
brindle keep going without you and stop when the work is actually done.
|
|
235
234
|
3. **Set `checks` in `.brindle/config.json`** (usually your full test suite) so no
|
|
236
235
|
branch merges red, and a `setup` if new worktrees need `npm install` or similar.
|
|
237
236
|
4. **Watch the sidebar, not every window.** It flags agents waiting on you (◆);
|
|
@@ -249,9 +248,9 @@ task to one area of the code with its own test command, and let the supervisor r
|
|
|
249
248
|
to `max_agents` workers at once. Use a cheaper profile (see [Cheap workers](#cheap-workers))
|
|
250
249
|
for mechanical edits and a Codex reviewer for a second model's opinion.
|
|
251
250
|
|
|
252
|
-
##
|
|
251
|
+
## Driving toward a verified goal
|
|
253
252
|
|
|
254
|
-
`brindle`
|
|
253
|
+
`brindle` drives the supervisor toward a verified goal. Tell it what we're building,
|
|
255
254
|
or put the goal in `.brindle/goals.md`, and it works like a project manager:
|
|
256
255
|
|
|
257
256
|
1. **Goal and milestones.** The goal is split into milestones, and each one has
|
|
@@ -277,7 +276,7 @@ or put the goal in `.brindle/goals.md`, and it works like a project manager:
|
|
|
277
276
|
approves it, or (`approved=false`, with feedback) sends it back for a
|
|
278
277
|
revision. For Claude workers brindle's PreToolUse hook refuses Edit, Write
|
|
279
278
|
and NotebookEdit until the plan is approved; other CLIs are only told to
|
|
280
|
-
wait.
|
|
279
|
+
wait. brindle doesn't count a worker waiting on approval as stalled, and
|
|
281
280
|
reminds the supervisor about plans awaiting a decision.
|
|
282
281
|
3. **Gated merges.** A branch merges only when everything is committed, a
|
|
283
282
|
reviewer agent has approved that exact commit, your pre-commit hooks pass,
|
|
@@ -321,7 +320,7 @@ for byte as you wrote it. Since `goals.md` may be committed, it stays free of
|
|
|
321
320
|
session data: no agent or session ids, check output, notes, usage or
|
|
322
321
|
questions ever go in. A status line is information only: a new session
|
|
323
322
|
starts every milestone as pending and re-runs the checks, never trusting it.
|
|
324
|
-
A session stops writing when it's handed over
|
|
323
|
+
A session stops writing when it's handed over or paused,
|
|
325
324
|
or once its goal was replaced from the chat. From a linked worktree, it writes
|
|
326
325
|
the `goals.md` it loaded (the main checkout's), never another session's. A
|
|
327
326
|
write that fails never fails the check.
|
|
@@ -331,15 +330,14 @@ A new session takes a hand-written `goals.md` over by moving it to its own
|
|
|
331
330
|
its deletion is left for you to commit or restore; brindle doesn't count it as
|
|
332
331
|
an uncommitted change, so it never blocks a merge or a milestone's check.
|
|
333
332
|
|
|
334
|
-
`brindle
|
|
335
|
-
|
|
336
|
-
|
|
337
|
-
starts without it.
|
|
333
|
+
`brindle goal` shows progress and `brindle goal check` runs the checks now.
|
|
334
|
+
Pause the session to stop it. The deprecated `--autopilot`/`--no-autopilot`
|
|
335
|
+
flags and the `autopilot` config key are ignored: goal driving is always on.
|
|
338
336
|
|
|
339
|
-
When your Claude usage reaches `usage_limit`,
|
|
337
|
+
When your Claude usage reaches `usage_limit`, brindle pauses for usage: its
|
|
340
338
|
running Claude workers stop (worktrees, branches, queued messages and sessions
|
|
341
339
|
are kept) and the sidebar says "paused for usage until <time>". Once the usage
|
|
342
|
-
window resets, brindle restarts those workers on its own,
|
|
340
|
+
window resets, brindle restarts those workers on its own, resumes goal driving
|
|
343
341
|
again and tells the supervisor what it resumed. Speaking to the supervisor
|
|
344
342
|
doesn't end the pause.
|
|
345
343
|
|
|
@@ -359,9 +357,9 @@ your own status line prints, so what you see doesn't change.
|
|
|
359
357
|
| `brindle new BRANCH [-b BASE] [-p PROMPT] [--no-agent]` | worktree + branch + a supervisor in it |
|
|
360
358
|
| `brindle continue [ID]` / `brindle -c` | resume a paused session (default: the most recent) |
|
|
361
359
|
| `brindle sessions` / `brindle prune` | list paused sessions / apply retention rules, remove merged or missing worktrees, close stale stopped agents whose work is merged or gone, and remove leftover tmux sessions |
|
|
362
|
-
| `brindle start [-p PROMPT] [--no-watch] [
|
|
360
|
+
| `brindle start [-p PROMPT] [--no-watch] [-b BRANCH] [-w PATH]` | the same, with options; `-b`/`-w` run it in that branch's worktree (created if needed, or the one you already made), which gets the repo's `.brindle` config |
|
|
363
361
|
| `brindle handover --to BRANCH\|PATH [-n NOTE]` | hand the session to a new supervisor there: goal and milestones, workers, queued tasks and your note move across; the old one is paused |
|
|
364
|
-
| `brindle
|
|
362
|
+
| `brindle goal [check]` | the goal's progress; run the checks now |
|
|
365
363
|
| `brindle audit-decision FINDING [-r RATIONALE]` | record acceptance of an audit finding with rationale so subsequent audits respect it |
|
|
366
364
|
| `brindle delegation [conservative\|balanced\|fast]` | how readily the supervisor delegates: fewest tokens, the default, or quickest |
|
|
367
365
|
| `brindle transfer [REPO] [--from SESSION] [-b BRANCH]` | move a scratch session's work into a real repo |
|
|
@@ -411,7 +409,7 @@ With no `WS` argument, commands act on the workspace you're in.
|
|
|
411
409
|
`brindle web` serves the current session to your browser: workspaces, agents
|
|
412
410
|
and their status, and the goal's milestones with each one's last check. Run
|
|
413
411
|
it from a terminal in the session's tmux session to show that session's
|
|
414
|
-
goal; elsewhere it shows the newest
|
|
412
|
+
goal; elsewhere it shows the newest session
|
|
415
413
|
in the repo. The page refreshes itself every couple of seconds.
|
|
416
414
|
|
|
417
415
|
It listens on `127.0.0.1` only, on a free port (or `--port N`), and prints a
|
|
@@ -443,7 +441,7 @@ knowing them helps when you tell the supervisor how to work.
|
|
|
443
441
|
| `approve_plan` | supervisor | approve a worker's plan, or send it back with feedback (`approved=false`) |
|
|
444
442
|
| `answer_permission` | supervisor | allow, deny or escalate (with a reason) a worker's permission ask before the person sees it |
|
|
445
443
|
| `complete_subagent` | supervisor | record the result of a `subagent`-profile task |
|
|
446
|
-
| `set_goal` / `get_progress` / `check_milestone` | supervisor |
|
|
444
|
+
| `set_goal` / `get_progress` / `check_milestone` | supervisor | the session's goal, its progress, and running the checks |
|
|
447
445
|
| `record_audit_decision` | supervisor | record acceptance of an audit finding with rationale so subsequent audits respect it |
|
|
448
446
|
| `who_is_working` | supervisor | other sessions in this repo: their goals and running tasks, optionally only those matching `paths` or a `query` |
|
|
449
447
|
| `record_decision` / `decisions` | supervisor (`decisions` also workers) | record a durable decision about the system / list the accepted ones that apply |
|
|
@@ -512,7 +510,7 @@ Set in `~/.brindle/config.json`, `.brindle/config.json` or `config.local.json`
|
|
|
512
510
|
take the task, its goal or the month over a limit (a task's cost is estimated
|
|
513
511
|
from this repo's finished tasks on that profile, else a typical token mix at
|
|
514
512
|
list prices), brindle skips that candidate and uses the next cheaper one. If
|
|
515
|
-
none fits it refuses the assign, says why and
|
|
513
|
+
none fits it refuses the assign, says why and asks you. A
|
|
516
514
|
profile you name yourself is never second-guessed. Running workers are
|
|
517
515
|
reported to the supervisor at 80% of `task_usd` and again at 100%; set
|
|
518
516
|
`"stop": true` to close a worker at 100% instead of only warning. Without the
|
|
@@ -601,7 +599,7 @@ brindle coordinator stop # close it and turn it off for this che
|
|
|
601
599
|
brindle keeps one place (`~/.brindle/quota.json`) that knows how close each
|
|
602
600
|
provider is to its subscription limit, and shows a note per provider that has
|
|
603
601
|
data (e.g. `Codex at 82% of its weekly limit, resets Thu 9:00am`) in `assign`
|
|
604
|
-
replies, `get_progress`, the sidebar's
|
|
602
|
+
replies, `get_progress`, the sidebar's goal block and `brindle doctor`.
|
|
605
603
|
It only uses what the CLIs write locally, and never reads a CLI's login
|
|
606
604
|
token or auth files or calls a provider's servers:
|
|
607
605
|
|
|
@@ -634,16 +632,16 @@ A linked git worktree you made yourself doesn't have the git-ignored parts of
|
|
|
634
632
|
}
|
|
635
633
|
```
|
|
636
634
|
|
|
637
|
-
|
|
635
|
+
Goal driving, merge gates and cleanup:
|
|
638
636
|
|
|
639
637
|
| Key | Default | |
|
|
640
638
|
|---|---|---|
|
|
641
|
-
| `autopilot` | `true` | start the supervisor with autopilot on |
|
|
642
639
|
| `claude_permission_mode` | `"default"` | the `--permission-mode` for Claude Code sessions whose profile sets no `permission_mode`; the default holds permission requests for the brindle app, set `"auto"` to opt into auto mode. Only read from `~/.brindle/config.json` |
|
|
643
640
|
| `auto_update` | `false` | run `uv tool upgrade brindle` in the background at session start when PyPI has a newer version; restart to use it. Can also be set in `~/.brindle/config.json` for every repo |
|
|
644
641
|
| `scoped_checks` | unset (acts as `"off"`) | `"infer"` narrows a bare pytest check (`pytest ...`, `python -m pytest ...`, `uv run pytest ...`, options only) on a worker's branch to the tests `graphify affected` reaches plus the touched test files, when the repo has a stored graph; it runs in full when the mapping fails or `conftest.py`, `pyproject.toml`, `setup.cfg`, `pytest.ini`, `tox.ini` or a lockfile changed. A pytest command that already names test paths (`pytest tests/`) and every other command always run in full. `"off"` narrows nothing. It becomes the default next release; until you set it, brindle tells you once |
|
|
642
|
+
| `scoped_checks_max_files` | `40` | with `scoped_checks`, a worker's branch whose selected tests exceed this many files runs only its own changed test files and each changed module's namesake tests (`tests/test_<module>.py`, `tests/test_<module>_*.py`) instead, and the check output carries a one-line note saying so. The session branch and the full run before a PR (`gates.pr_checked`) always run everything. `0`: no cap |
|
|
645
643
|
| `checks` | `[]` | commands (plain strings) that must pass in a worker's branch before it merges. On a worker's branch every check also gets `BRINDLE_CHANGED_FILES` (the files changed since the branch left its base, one repo-relative path per line) so a script can narrow itself; the supervisor's checkout always runs every check in full (see `scoped_checks` and `brindle/check_scope.py`) |
|
|
646
|
-
| `review` |
|
|
644
|
+
| `review` | required for sessions | require a reviewer's approval before merging |
|
|
647
645
|
| `reviewer` | `"reviewer"` | the agent profile that reviews (a Codex profile gives a second model's view) |
|
|
648
646
|
| `review_profile` | none | force `request_review`'s profile, skipping its automatic cross-model pick (see below) |
|
|
649
647
|
| `pre_commit` | `true` | run [pre-commit](https://pre-commit.com) over the branch, if the repo uses it |
|
|
@@ -651,7 +649,7 @@ Autopilot, merge gates and cleanup:
|
|
|
651
649
|
| `check_timeout` | `900` | seconds each check may take |
|
|
652
650
|
| `check_concurrency` | half the logical cores, at least 2, at most 6 and 2 GB of RAM per run | check runs at once on this machine, across branches; the rest wait their turn (`0`: no cap). A run far past its last duration is reported to the supervisor |
|
|
653
651
|
| `fable_escalation` | on, unless the worker runs on a cloud account (Bedrock, Vertex or Foundry) | retry a failed heavy task once on `developer-heavy` before handing it to you; `true` turns it on for a cloud account, `false` off |
|
|
654
|
-
| `usage_limit` | `90` |
|
|
652
|
+
| `usage_limit` | `90` | brindle stops pushing on at this % of your Claude usage limit |
|
|
655
653
|
| `limit_cooldown_minutes` | `300` for Antigravity | how long a provider that hit its limit counts as unavailable |
|
|
656
654
|
| `graphify` | if the graph is there | point agents at the repo's [graphify](https://github.com/safishamsi/graphify) code map (`false` turns it off) |
|
|
657
655
|
| `graphify_auto_refresh` | `true` | automatically rebuild untracked code maps; `false` disables rebuilds while keeping queries available |
|
|
@@ -783,7 +781,7 @@ probe here; no Claude CLI or synthetic generation request is used for these chec
|
|
|
783
781
|
The real first worker task is sent once. A detached helper inspects its current
|
|
784
782
|
pane five seconds after delivery, so readiness delays do not use up that window.
|
|
785
783
|
Fresh interactive Antigravity chats without a task still receive their profile,
|
|
786
|
-
tool instructions and
|
|
784
|
+
tool instructions and goal context once after readiness; worker
|
|
787
785
|
instructions accompany the actual task in the same delivery.
|
|
788
786
|
For argv tasks, the timer starts at prompt submission or observed CLI readiness,
|
|
789
787
|
after trust/startup handling; readiness timeout does not establish delivery or
|
|
@@ -1499,7 +1497,7 @@ never handed to a worker: the supervisor gets them as "needs you".
|
|
|
1499
1497
|
|
|
1500
1498
|
**The completion audit.** A milestone's check can pass without the milestone being
|
|
1501
1499
|
done: the test asserts too little, or checks the wrong thing. So when every milestone
|
|
1502
|
-
passes,
|
|
1500
|
+
passes, brindle doesn't call the goal reached yet. A read-only reviewer audits everything since the goal was set
|
|
1503
1501
|
against the goal and each milestone, looking for what's missing, stubbed or weakly
|
|
1504
1502
|
tested. Only its approval of that commit finishes the goal. If it finds gaps, the
|
|
1505
1503
|
supervisor gets them as a message and keeps working until a fresh audit approves.
|
|
@@ -1550,7 +1548,7 @@ commands around the team's list. `~/.brindle/config.json` holds your own default
|
|
|
1550
1548
|
for every repo (any key except the command lists, like `delegation` or
|
|
1551
1549
|
`sidebar`); a repo's two files override it. With brindle Pro or higher, those
|
|
1552
1550
|
preferences (`delegation`, `sidebar`, `message_delivery`, `pr_footer`,
|
|
1553
|
-
`delete_merged_branches`, `max_agents`, `
|
|
1551
|
+
`delete_merged_branches`, `max_agents`, `plan_first`, `stale_after`,
|
|
1554
1552
|
`usage_limit`, `review_rounds`) sync across your machines (last change wins; nothing
|
|
1555
1553
|
else in the file leaves the machine); `brindle account sync` syncs now.
|
|
1556
1554
|
|
|
@@ -1658,16 +1656,16 @@ Your org's `deny_personal_keys` (Enterprise) still applies on top of `auth`.
|
|
|
1658
1656
|
|
|
1659
1657
|
#### Running unattended on API keys
|
|
1660
1658
|
|
|
1661
|
-
When running long, unattended
|
|
1659
|
+
When running long, unattended sessions with API keys:
|
|
1662
1660
|
|
|
1663
1661
|
- **Quota, token and cost measurements per provider**:
|
|
1664
|
-
-
|
|
1662
|
+
- brindle's automatic usage pause (`usage_limit: 90`) relies on the percentage numbers emitted by Claude Code's status line under subscription plans. Under API-key billing (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `GEMINI_API_KEY`), the CLIs report no subscription percentage, so subscription usage pause thresholds do not trigger; manage spend via provider account budget limits or brindle budget policies.
|
|
1665
1663
|
- Token and cost tracking: Token counts and cost estimates are captured from Claude and Codex transcripts. Antigravity (`agy`) currently has **no token/cost measurements**, so budget policies and cost tracking cannot monitor Antigravity spend; enforce usage caps directly in Google AI Studio or your Google Cloud project billing settings.
|
|
1666
1664
|
- **Runtime auth failure handling and provider differences**:
|
|
1667
|
-
- **Claude and Antigravity**: Runtime authentication failures (such as 401/403, revoked API keys, or credit/billing errors) are intercepted via StopFailure and exit hooks. Brindle marks the worker as failed, alerts the supervisor, invalidates cached sign-in checks, and pauses
|
|
1668
|
-
- **Codex (unsupported runtime auth detection)**: Codex CLI turn completion (`agent-turn-complete`) does not classify auth errors at runtime. If an OpenAI API key expires or is rejected, Codex workers will not automatically transition to `failed` or trigger
|
|
1665
|
+
- **Claude and Antigravity**: Runtime authentication failures (such as 401/403, revoked API keys, or credit/billing errors) are intercepted via StopFailure and exit hooks. Brindle marks the worker as failed, alerts the supervisor, invalidates cached sign-in checks, and pauses goal driving with a `Credentials need attention` notice rather than cycling idle nudges.
|
|
1666
|
+
- **Codex (unsupported runtime auth detection)**: Codex CLI turn completion (`agent-turn-complete`) does not classify auth errors at runtime. If an OpenAI API key expires or is rejected, Codex workers will not automatically transition to `failed` or trigger a credential pause. When running Codex workers unattended, monitor their progress via `brindle watch` or check tmux panes; if an auth failure occurs, manually restart or rotate keys.
|
|
1669
1667
|
- **Supervisor credential refresh and key rotation**:
|
|
1670
|
-
- **Recovering failed/blocked workers**: When an agent fails on an expired key or
|
|
1668
|
+
- **Recovering failed/blocked workers**: When an agent fails on an expired key or brindle pauses with a credential error, export the fresh key (or update `brindle keys set`), then run `brindle continue`. This unpauses the session and relaunches paused/failed agents with the new credentials. The replacement credentials are automatically synced to the session credential store, dynamically refreshed by the supervisor's launcher on all subsequent worker delegations (`assign`/`handoff`).
|
|
1671
1669
|
- **Proactive rotation in live sessions**: For Claude keys rotated outside brindle, set Claude Code's `apiKeyHelper` in `~/.claude/settings.json` or the project's `.claude/settings.json`. Claude Code re-runs the helper every 5 minutes (configurable with `CLAUDE_CODE_API_KEY_HELPER_TTL_MS`) and after a 401, so running workers pick up the new key. Don't also export `ANTHROPIC_API_KEY` or set a Claude key with `brindle keys set`, since an environment key takes precedence over the helper. Codex and Antigravity keys still need workers stopped, then `brindle keys set`, then `brindle continue`. Exporting a new key while an active session is healthy does not restart running workers or the supervisor (running `brindle continue` on an active healthy session simply reattaches to tmux without restarting workers). To proactively rotate credentials for the entire session and all running workers: run `brindle stop` -> update credentials in your environment or via `brindle keys set` -> run `brindle continue`.
|
|
1672
1670
|
- **Precedence**: Newly stored keys (`brindle keys set`) or session credentials updated during recovery take effect dynamically for all new panes, overriding stale launcher environments. For profile-level overrides, an explicit profile key (`env.NAME`) takes precedence over default provider credentials.
|
|
1673
1671
|
- **Antigravity permissions prerequisite**: Antigravity (`agy`) does not support approving shell commands solely via external hooks. brindle therefore runs agy workers with `--dangerously-skip-permissions` and its own hook as the gate, applying their profile's `allowed_tools` and denying the rest (see Permissions). Under an explicit `agy_approvals: "prompt"` they instead stall on approval prompts for anything `settings.json` doesn't allow, their profile's `allowed_tools` included.
|
|
@@ -2116,7 +2114,7 @@ To run workers on your own Bedrock, Vertex, Codex or Antigravity account, see
|
|
|
2116
2114
|
Codex agents report status through Codex's `notify` hook (brindle passes
|
|
2117
2115
|
`-c notify=[...]` at launch, leaving your own Codex config alone): a completed turn
|
|
2118
2116
|
marks the agent idle and delivers any queued message. Codex has no Stop hook, so for
|
|
2119
|
-
|
|
2117
|
+
a supervisor on Codex that turn end is also where brindle tells it to keep
|
|
2120
2118
|
going, with the same limit on reminders as Claude Code.
|
|
2121
2119
|
|
|
2122
2120
|
### Running workers on your own cloud
|
|
@@ -2335,7 +2333,7 @@ model's review) and the supervisor can hand it tasks.
|
|
|
2335
2333
|
|
|
2336
2334
|
`agy` has no command-line options for hooks, MCP servers or instructions, so brindle
|
|
2337
2335
|
adds three files to the checkout's `.agents/` folder: `mcp_config.json` (brindle's tools),
|
|
2338
|
-
`hooks.json` (status, messages,
|
|
2336
|
+
`hooks.json` (status, messages, goal driving) and `rules/brindle.md`. They're listed in
|
|
2339
2337
|
`.git/info/exclude`, so they never show up in `git status`. brindle adds to these files if
|
|
2340
2338
|
you already have them, and won't change one that's committed. Each agent's first
|
|
2341
2339
|
message is a short warm-up with its instructions, because `agy` connects MCP servers
|
|
@@ -2378,7 +2376,7 @@ add rules to that file yourself, for example:
|
|
|
2378
2376
|
- **Messages wake a supervisor through brindle's own hook.** Your interactive
|
|
2379
2377
|
Claude Code chat runs a small background waiter (an `asyncRewake` hook started
|
|
2380
2378
|
at `SessionStart` and at every `Stop`). When a worker reports, or brindle has a
|
|
2381
|
-
"new messages" notice or
|
|
2379
|
+
"new messages" notice or a goal nudge, the waiter hands it over and Claude
|
|
2382
2380
|
Code wakes the model with it: an idle chat starts a new turn, a busy one gets it
|
|
2383
2381
|
between steps, and the chat shows one "Stop hook feedback" line. Each message
|
|
2384
2382
|
is wrapped as `<brindle-msg nonce="..." from="developer agent w1">...</brindle-msg>`
|