@chankov/agent-fleet 0.0.6 → 0.0.7
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.
- package/.claude/commands/set-hermes-watchdog.md +6 -0
- package/.opencode/commands/af-set-hermes-watchdog.md +6 -0
- package/.pi/harnesses/agent-hub/README.md +75 -16
- package/.pi/harnesses/agent-hub/artifacts.js +56 -2
- package/.pi/harnesses/agent-hub/assertion-ledger.js +58 -0
- package/.pi/harnesses/agent-hub/context-window.js +106 -0
- package/.pi/harnesses/agent-hub/delegate.ts +58 -9
- package/.pi/harnesses/agent-hub/drift-watchdog.js +54 -8
- package/.pi/harnesses/agent-hub/helpers.ts +34 -0
- package/.pi/harnesses/agent-hub/index.ts +723 -159
- package/.pi/harnesses/agent-hub/monitor-invoke-journal.ts +131 -0
- package/.pi/harnesses/agent-hub/monitor-invoke.ts +92 -0
- package/.pi/harnesses/agent-hub/monitor-lifecycle.ts +2 -2
- package/.pi/harnesses/agent-hub/monitor-session-bridge.ts +257 -11
- package/.pi/harnesses/agent-hub/package.json +1 -1
- package/.pi/harnesses/agent-hub/provider-semaphore.js +138 -0
- package/.pi/harnesses/agent-hub/return-contract.js +78 -0
- package/.pi/harnesses/agent-hub/return-extract.js +75 -0
- package/.pi/harnesses/agent-hub/run-budget.js +29 -3
- package/.pi/harnesses/agent-hub/session-health.js +110 -0
- package/.pi/harnesses/agent-hub/spawn.ts +109 -1
- package/.pi/harnesses/agent-hub/spawned-peers.js +74 -0
- package/.pi/harnesses/coms/README.md +4 -3
- package/.pi/harnesses/coms/index.ts +80 -58
- package/.pi/harnesses/coms/package.json +1 -1
- package/.pi/harnesses/damage-control-continue/package.json +1 -1
- package/.pi/harnesses/lib/coms-registry-entry.ts +94 -0
- package/.pi/harnesses/lib/herdr-client.ts +7 -2
- package/.pi/harnesses/lib/herdr-presence.ts +141 -30
- package/.pi/harnesses/lib/hermes-monitor-events.ts +136 -0
- package/.pi/harnesses/lib/hermes-monitor-model.ts +9 -0
- package/.pi/harnesses/lib/hermes-monitor-registry.ts +1 -1
- package/.pi/harnesses/lib/hermes-monitor-socket.ts +47 -2
- package/.pi/prompts/af-set-hermes-watchdog.md +6 -0
- package/.versions/0.0.7/.claude/commands/build.md +44 -0
- package/.versions/0.0.7/.claude/commands/code-simplify.md +22 -0
- package/.versions/0.0.7/.claude/commands/design-agent.md +14 -0
- package/.versions/0.0.7/.claude/commands/doctor-agent-fleet.md +17 -0
- package/.versions/0.0.7/.claude/commands/orchestrate.md +112 -0
- package/.versions/0.0.7/.claude/commands/plan.md +16 -0
- package/.versions/0.0.7/.claude/commands/prime.md +22 -0
- package/.versions/0.0.7/.claude/commands/review.md +16 -0
- package/.versions/0.0.7/.claude/commands/set-hermes-telegram.md +15 -0
- package/.versions/0.0.7/.claude/commands/set-hermes-watchdog.md +6 -0
- package/.versions/0.0.7/.claude/commands/setup-agent-fleet.md +25 -0
- package/.versions/0.0.7/.claude/commands/ship.md +72 -0
- package/.versions/0.0.7/.claude/commands/spec.md +15 -0
- package/.versions/0.0.7/.claude/commands/test.md +19 -0
- package/.versions/0.0.7/.claude/commands/webperf.md +32 -0
- package/.versions/0.0.7/.claude/orchestrate-teams.yaml +24 -0
- package/.versions/0.0.7/.opencode/commands/af-build.md +17 -0
- package/.versions/0.0.7/.opencode/commands/af-code-simplify.md +16 -0
- package/.versions/0.0.7/.opencode/commands/af-compound.md +16 -0
- package/.versions/0.0.7/.opencode/commands/af-design-agent.md +15 -0
- package/.versions/0.0.7/.opencode/commands/af-doctor-agent-fleet.md +17 -0
- package/.versions/0.0.7/.opencode/commands/af-orchestrate.md +112 -0
- package/.versions/0.0.7/.opencode/commands/af-plan.md +16 -0
- package/.versions/0.0.7/.opencode/commands/af-prime.md +22 -0
- package/.versions/0.0.7/.opencode/commands/af-review.md +15 -0
- package/.versions/0.0.7/.opencode/commands/af-set-hermes-telegram.md +15 -0
- package/.versions/0.0.7/.opencode/commands/af-set-hermes-watchdog.md +6 -0
- package/.versions/0.0.7/.opencode/commands/af-setup-agent-fleet.md +15 -0
- package/.versions/0.0.7/.opencode/commands/af-ship.md +16 -0
- package/.versions/0.0.7/.opencode/commands/af-spec.md +16 -0
- package/.versions/0.0.7/.opencode/commands/af-test.md +21 -0
- package/.versions/0.0.7/.opencode/commands/af-webperf.md +32 -0
- package/.versions/0.0.7/.opencode/orchestrate-teams.yaml +24 -0
- package/.versions/0.0.7/.pi/agents/dispatch-policy.yaml +32 -0
- package/.versions/0.0.7/.pi/agents/model-profiles.yaml +20 -0
- package/.versions/0.0.7/.pi/agents/peers.yaml +146 -0
- package/.versions/0.0.7/.pi/agents/teams.yaml +39 -0
- package/.versions/0.0.7/.pi/damage-control-rules.yaml +297 -0
- package/.versions/0.0.7/.pi/extensions/agent-fleet-update-check/README.md +58 -0
- package/.versions/0.0.7/.pi/extensions/agent-fleet-update-check/index.ts +161 -0
- package/.versions/0.0.7/.pi/extensions/agent-fleet-update-check/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/btw/README.md +118 -0
- package/.versions/0.0.7/.pi/extensions/btw/index.ts +917 -0
- package/.versions/0.0.7/.pi/extensions/btw/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/btw/state.js +179 -0
- package/.versions/0.0.7/.pi/extensions/chrome-devtools-mcp/README.md +58 -0
- package/.versions/0.0.7/.pi/extensions/chrome-devtools-mcp/index.ts +93 -0
- package/.versions/0.0.7/.pi/extensions/chrome-devtools-mcp/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/compact-and-continue/README.md +42 -0
- package/.versions/0.0.7/.pi/extensions/compact-and-continue/index.ts +120 -0
- package/.versions/0.0.7/.pi/extensions/compact-and-continue/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/mcp-bridge/README.md +46 -0
- package/.versions/0.0.7/.pi/extensions/mcp-bridge/index.ts +206 -0
- package/.versions/0.0.7/.pi/extensions/mcp-bridge/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/package-lock.json +1143 -0
- package/.versions/0.0.7/.pi/extensions/package.json +9 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/README.md +201 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/config.ts +198 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/examples/stt.json +35 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/index.ts +248 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/package.json +6 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/providers.ts +214 -0
- package/.versions/0.0.7/.pi/extensions/pi-voice-stt/recorder.ts +158 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/README.md +857 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/access-approval.ts +135 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/artifacts.js +146 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/assertion-ledger.js +58 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/backend-policy.js +153 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/context-window.js +106 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/delegate.ts +435 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/drift-watchdog.js +211 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/evidence-rules.js +96 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/footer.ts +29 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/helpers.ts +248 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/index.ts +8609 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-control.ts +9 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-invoke-journal.ts +131 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-invoke.ts +92 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-lifecycle.ts +13 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-publisher.ts +59 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-recovery.ts +3 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-runtime.ts +3 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/monitor-session-bridge.ts +257 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/package.json +7 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/provider-semaphore.js +138 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/research-retention.js +34 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/research-watchdog.ts +40 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/return-contract.js +293 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/return-extract.js +75 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/run-budget.js +181 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/safety-routing.ts +22 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/scope-gate.js +103 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/session-health.js +110 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/spawn.ts +354 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/spawned-peers.js +74 -0
- package/.versions/0.0.7/.pi/harnesses/agent-hub/version.ts +18 -0
- package/.versions/0.0.7/.pi/harnesses/ask-user-remote/README.md +19 -0
- package/.versions/0.0.7/.pi/harnesses/ask-user-remote/index.ts +321 -0
- package/.versions/0.0.7/.pi/harnesses/ask-user-remote/race-core.js +106 -0
- package/.versions/0.0.7/.pi/harnesses/coms/README.md +93 -0
- package/.versions/0.0.7/.pi/harnesses/coms/index.ts +1767 -0
- package/.versions/0.0.7/.pi/harnesses/coms/package.json +7 -0
- package/.versions/0.0.7/.pi/harnesses/coms/version.ts +18 -0
- package/.versions/0.0.7/.pi/harnesses/damage-control-continue/README.md +117 -0
- package/.versions/0.0.7/.pi/harnesses/damage-control-continue/index.ts +536 -0
- package/.versions/0.0.7/.pi/harnesses/damage-control-continue/package.json +7 -0
- package/.versions/0.0.7/.pi/harnesses/damage-control-continue/version.ts +18 -0
- package/.versions/0.0.7/.pi/harnesses/lib/coms-registry-entry.ts +94 -0
- package/.versions/0.0.7/.pi/harnesses/lib/damage-control-shared.ts +197 -0
- package/.versions/0.0.7/.pi/harnesses/lib/herdr-client.ts +463 -0
- package/.versions/0.0.7/.pi/harnesses/lib/herdr-presence.ts +399 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-events.ts +136 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-herdr.ts +28 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-model.ts +18 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-registry.ts +6 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-socket.ts +163 -0
- package/.versions/0.0.7/.pi/harnesses/lib/hermes-monitor-store.ts +31 -0
- package/.versions/0.0.7/.pi/harnesses/package-lock.json +35 -0
- package/.versions/0.0.7/.pi/harnesses/package.json +9 -0
- package/.versions/0.0.7/.pi/prompts/af-build.md +24 -0
- package/.versions/0.0.7/.pi/prompts/af-code-simplify.md +22 -0
- package/.versions/0.0.7/.pi/prompts/af-doctor-agent-fleet.md +15 -0
- package/.versions/0.0.7/.pi/prompts/af-plan.md +16 -0
- package/.versions/0.0.7/.pi/prompts/af-review.md +16 -0
- package/.versions/0.0.7/.pi/prompts/af-set-hermes-telegram.md +15 -0
- package/.versions/0.0.7/.pi/prompts/af-set-hermes-watchdog.md +6 -0
- package/.versions/0.0.7/.pi/prompts/af-setup-agent-fleet.md +27 -0
- package/.versions/0.0.7/.pi/prompts/af-ship.md +17 -0
- package/.versions/0.0.7/.pi/prompts/af-spec.md +15 -0
- package/.versions/0.0.7/.pi/prompts/af-test.md +19 -0
- package/.versions/0.0.7/.pi/skills/bowser/SKILL.md +145 -0
- package/.versions/0.0.7/.version +1 -0
- package/.versions/0.0.7/agents/architect.md +23 -0
- package/.versions/0.0.7/agents/bowser.md +26 -0
- package/.versions/0.0.7/agents/builder.md +53 -0
- package/.versions/0.0.7/agents/code-reviewer.md +205 -0
- package/.versions/0.0.7/agents/deep-researcher.md +31 -0
- package/.versions/0.0.7/agents/documenter.md +18 -0
- package/.versions/0.0.7/agents/orchestrator.md +65 -0
- package/.versions/0.0.7/agents/plan-reviewer.md +60 -0
- package/.versions/0.0.7/agents/planner.md +69 -0
- package/.versions/0.0.7/agents/releaser.md +24 -0
- package/.versions/0.0.7/agents/researcher.md +28 -0
- package/.versions/0.0.7/agents/security-auditor.md +166 -0
- package/.versions/0.0.7/agents/test-engineer.md +151 -0
- package/.versions/0.0.7/agents/web-debugger.md +42 -0
- package/.versions/0.0.7/agents/web-performance-auditor.md +192 -0
- package/.versions/0.0.7/codex/CONDUCTOR.md +148 -0
- package/.versions/0.0.7/docs/ARCHITECTURE.md +262 -0
- package/.versions/0.0.7/docs/agent-fleet-setup.md +366 -0
- package/.versions/0.0.7/docs/claude-code-coms-bridge.md +109 -0
- package/.versions/0.0.7/docs/codex-remote-conductor.md +170 -0
- package/.versions/0.0.7/docs/coms-hermes-bridge.md +189 -0
- package/.versions/0.0.7/docs/getting-started.md +231 -0
- package/.versions/0.0.7/docs/npm-install.md +337 -0
- package/.versions/0.0.7/docs/opencode-setup.md +249 -0
- package/.versions/0.0.7/docs/pi-extensions.md +293 -0
- package/.versions/0.0.7/docs/pi-setup.md +476 -0
- package/.versions/0.0.7/docs/skill-anatomy.md +150 -0
- package/.versions/0.0.7/hermes/README.md +347 -0
- package/.versions/0.0.7/hermes/desktop-plugins/agent-fleet-herdr/plugin.js +1291 -0
- package/.versions/0.0.7/hermes/desktop-plugins/agent-fleet-herdr/presentation.js +672 -0
- package/.versions/0.0.7/hermes/desktop-plugins/agent-fleet-monitor/plugin.js +19 -0
- package/.versions/0.0.7/hermes/desktop-plugins/agent-fleet-monitor/state.js +10 -0
- package/.versions/0.0.7/hermes/gates/gate-o-capability.md +8 -0
- package/.versions/0.0.7/hermes/gates/gate-o-validator-fixtures.json +1 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/activity.py +770 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/coms_registry.py +134 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/coms_send.py +350 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/herdr_source.py +193 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/manifest.json +4 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/plugin_api.py +444 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/sessions.py +205 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/tasks.py +300 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/dashboard/watch.py +819 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-herdr/plugin.yaml +8 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-monitor/dashboard/adapter.py +139 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-monitor/dashboard/manifest.json +11 -0
- package/.versions/0.0.7/hermes/plugins/agent-fleet-monitor/dashboard/plugin_api.py +50 -0
- package/.versions/0.0.7/hermes/skills/hub-conductor/SKILL.md +84 -0
- package/.versions/0.0.7/hermes/skills/hub-liaison/SKILL.md +84 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/SKILL.md +76 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/manifest.json +1 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog.py +249 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_actions.py +124 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_contract.py +76 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_delivery.py +10 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_engine.py +128 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_judgment.py +41 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_recovery.py +103 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_state.py +163 -0
- package/.versions/0.0.7/hermes/skills/hub-watchdog/scripts/watchdog_transport.py +138 -0
- package/.versions/0.0.7/hermes/watchdog-tests/__init__.py +0 -0
- package/.versions/0.0.7/hooks/SIMPLIFY-IGNORE.md +90 -0
- package/.versions/0.0.7/hooks/coms-stop-hook.mjs +84 -0
- package/.versions/0.0.7/hooks/hooks.json +14 -0
- package/.versions/0.0.7/hooks/session-start.sh +74 -0
- package/.versions/0.0.7/hooks/simplify-ignore.sh +302 -0
- package/.versions/0.0.7/justfile +245 -0
- package/.versions/0.0.7/references/accessibility-checklist.md +160 -0
- package/.versions/0.0.7/references/definition-of-done.md +67 -0
- package/.versions/0.0.7/references/fleet-coordination-patterns.md +151 -0
- package/.versions/0.0.7/references/observability-checklist.md +91 -0
- package/.versions/0.0.7/references/orchestration-patterns.md +370 -0
- package/.versions/0.0.7/references/performance-checklist.md +153 -0
- package/.versions/0.0.7/references/prompting-patterns.md +380 -0
- package/.versions/0.0.7/references/security-checklist.md +179 -0
- package/.versions/0.0.7/references/testing-patterns.md +236 -0
- package/.versions/0.0.7/scripts/codex-conductor.ts +41 -0
- package/.versions/0.0.7/scripts/codex-remote-control.ts +167 -0
- package/.versions/0.0.7/scripts/coms-claude-bridge.ts +401 -0
- package/.versions/0.0.7/scripts/coms-cli.ts +484 -0
- package/.versions/0.0.7/scripts/coms-hermes-bridge.ts +413 -0
- package/.versions/0.0.7/scripts/fleet.ts +149 -0
- package/.versions/0.0.7/scripts/hermes-monitor-capability-runner.ts +25 -0
- package/.versions/0.0.7/scripts/hermes-monitor-scenario.ts +474 -0
- package/.versions/0.0.7/scripts/hermes-watchdog-capability-probe.ts +45 -0
- package/.versions/0.0.7/scripts/install-hermes-plugin.sh +180 -0
- package/.versions/0.0.7/scripts/lib/claude-bridge-core.ts +165 -0
- package/.versions/0.0.7/scripts/lib/codex-conductor.ts +88 -0
- package/.versions/0.0.7/scripts/lib/codex-remote-control.ts +480 -0
- package/.versions/0.0.7/scripts/lib/coms-envelope.ts +582 -0
- package/.versions/0.0.7/scripts/lib/fake-herdr-server.ts +88 -0
- package/.versions/0.0.7/scripts/lib/fleet-command.ts +109 -0
- package/.versions/0.0.7/scripts/lib/herdr-layout.ts +263 -0
- package/.versions/0.0.7/scripts/lib/hermes-bridge-core.ts +241 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-events.ts +1 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-herdr.ts +2 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-model.ts +3 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-registry.ts +2 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-socket.ts +2 -0
- package/.versions/0.0.7/scripts/lib/hermes-monitor-store.ts +2 -0
- package/.versions/0.0.7/scripts/lib/monitor-env.ts +120 -0
- package/.versions/0.0.7/scripts/lib/persona-banner.ts +62 -0
- package/.versions/0.0.7/scripts/lib/spawn-stagger.ts +101 -0
- package/.versions/0.0.7/scripts/lib/team-project.ts +191 -0
- package/.versions/0.0.7/scripts/lib/team-snapshot.ts +147 -0
- package/.versions/0.0.7/scripts/peer-banner.ts +40 -0
- package/.versions/0.0.7/scripts/team-snapshot.ts +237 -0
- package/.versions/0.0.7/scripts/team-up.ts +360 -0
- package/.versions/0.0.7/skills/_internal/grilling.md +11 -0
- package/.versions/0.0.7/skills/browser-testing-with-devtools/SKILL.md +352 -0
- package/.versions/0.0.7/skills/code-review-and-quality/SKILL.md +381 -0
- package/.versions/0.0.7/skills/compound-learning/SKILL.md +137 -0
- package/.versions/0.0.7/skills/context-engineering/SKILL.md +291 -0
- package/.versions/0.0.7/skills/deprecation-and-migration/SKILL.md +206 -0
- package/.versions/0.0.7/skills/designing-agents/SKILL.md +394 -0
- package/.versions/0.0.7/skills/designing-agents/pi-harness-authoring.md +222 -0
- package/.versions/0.0.7/skills/frontend-ui-engineering/SKILL.md +328 -0
- package/.versions/0.0.7/skills/git-workflow-and-versioning/SKILL.md +316 -0
- package/.versions/0.0.7/skills/guided-workspace-setup/SKILL.md +544 -0
- package/.versions/0.0.7/skills/guided-workspace-setup/companion-manifest.json +35 -0
- package/.versions/0.0.7/skills/idea-refine/SKILL.md +182 -0
- package/.versions/0.0.7/skills/idea-refine/examples.md +238 -0
- package/.versions/0.0.7/skills/idea-refine/frameworks.md +99 -0
- package/.versions/0.0.7/skills/idea-refine/refinement-criteria.md +113 -0
- package/.versions/0.0.7/skills/idea-refine/scripts/idea-refine.sh +15 -0
- package/.versions/0.0.7/skills/incremental-implementation/SKILL.md +287 -0
- package/.versions/0.0.7/skills/interview-me/SKILL.md +226 -0
- package/.versions/0.0.7/skills/orchestration-verification/SKILL.md +181 -0
- package/.versions/0.0.7/skills/peer-coms/SKILL.md +90 -0
- package/.versions/0.0.7/skills/performance-optimization/SKILL.md +350 -0
- package/.versions/0.0.7/skills/planning-and-task-breakdown/SKILL.md +236 -0
- package/.versions/0.0.7/skills/security-and-hardening/SKILL.md +461 -0
- package/.versions/0.0.7/skills/spec-driven-development/SKILL.md +245 -0
- package/.versions/0.0.7/skills/using-agent-skills/SKILL.md +199 -0
- package/.versions/0.0.7/systemd/user/agent-fleet-codex-remote-control.service.in +23 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/LICENSE +21 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/README.md +11 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/api-and-interface-design/SKILL.md +294 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/browser-testing-with-devtools/SKILL.md +317 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/ci-cd-and-automation/SKILL.md +390 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/code-review-and-quality/SKILL.md +396 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/code-simplification/SKILL.md +331 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/context-engineering/SKILL.md +289 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/debugging-and-error-recovery/SKILL.md +300 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/deprecation-and-migration/SKILL.md +247 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/documentation-and-adrs/SKILL.md +278 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/doubt-driven-development/SKILL.md +243 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/frontend-ui-engineering/SKILL.md +328 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/git-workflow-and-versioning/SKILL.md +355 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/idea-refine/SKILL.md +178 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/idea-refine/examples.md +238 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/idea-refine/frameworks.md +99 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/idea-refine/refinement-criteria.md +113 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/idea-refine/scripts/idea-refine.sh +15 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/incremental-implementation/SKILL.md +249 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/interview-me/SKILL.md +225 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/observability-and-instrumentation/SKILL.md +203 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/performance-optimization/SKILL.md +350 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/planning-and-task-breakdown/SKILL.md +234 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/security-and-hardening/SKILL.md +467 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/shipping-and-launch/SKILL.md +310 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/source-driven-development/SKILL.md +194 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/spec-driven-development/SKILL.md +206 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/test-driven-development/SKILL.md +383 -0
- package/.versions/0.0.7/vendor/agent-skills-upstream/skills/using-agent-skills/SKILL.md +191 -0
- package/CHANGELOG.md +134 -0
- package/README.md +175 -18
- package/agents/code-reviewer.md +2 -2
- package/agents/orchestrator.md +11 -0
- package/agents/plan-reviewer.md +1 -1
- package/agents/planner.md +1 -1
- package/agents/security-auditor.md +1 -1
- package/bin/cli.js +40 -1
- package/bin/lib/hermes-profile-artifact.js +144 -0
- package/bin/lib/set-hermes-watchdog.js +173 -0
- package/bin/set-hermes-watchdog.js +26 -0
- package/docs/ARCHITECTURE.md +64 -12
- package/docs/agent-fleet-setup.md +2 -2
- package/docs/claude-code-coms-bridge.md +7 -1
- package/docs/coms-hermes-bridge.md +10 -0
- package/docs/getting-started.md +48 -0
- package/docs/hermes-watchdog-supervisor.md +140 -0
- package/hermes/README.md +135 -19
- package/hermes/desktop-plugins/agent-fleet-herdr/plugin.js +1291 -0
- package/hermes/desktop-plugins/agent-fleet-herdr/presentation.js +672 -0
- package/hermes/desktop-plugins/agent-fleet-monitor/plugin.js +19 -0
- package/hermes/desktop-plugins/agent-fleet-monitor/state.js +10 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/activity.py +770 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/coms_registry.py +134 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/coms_send.py +350 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/herdr_source.py +193 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/manifest.json +4 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/plugin_api.py +444 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/sessions.py +205 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/tasks.py +300 -0
- package/hermes/plugins/agent-fleet-herdr/dashboard/watch.py +819 -0
- package/hermes/plugins/agent-fleet-herdr/plugin.yaml +8 -0
- package/hermes/plugins/agent-fleet-monitor/dashboard/adapter.py +139 -0
- package/hermes/plugins/agent-fleet-monitor/dashboard/dist/index.js +10 -0
- package/hermes/plugins/agent-fleet-monitor/dashboard/manifest.json +11 -0
- package/hermes/plugins/agent-fleet-monitor/dashboard/plugin_api.py +50 -0
- package/hermes/skills/hub-watchdog/SKILL.md +76 -0
- package/hermes/skills/hub-watchdog/manifest.json +1 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog.py +249 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_actions.py +124 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_contract.py +76 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_delivery.py +10 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_engine.py +128 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_judgment.py +41 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_recovery.py +103 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_state.py +163 -0
- package/hermes/skills/hub-watchdog/scripts/watchdog_transport.py +138 -0
- package/hooks/coms-stop-hook.mjs +12 -1
- package/package.json +11 -2
- package/scripts/coms-claude-bridge.ts +154 -39
- package/scripts/fleet.ts +9 -1
- package/scripts/hermes-watchdog-capability-probe.ts +45 -0
- package/scripts/lib/claude-bridge-core.ts +65 -0
- package/scripts/lib/coms-envelope.ts +44 -1
- package/scripts/lib/hermes-monitor-events.ts +1 -0
- package/scripts/lib/monitor-env.ts +120 -0
- package/scripts/team-snapshot.ts +1 -1
- package/scripts/team-up.ts +10 -1
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-cap.test.py +0 -15
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-discovery.test.py +0 -19
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-envelope.test.py +0 -13
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-forged-owner.test.py +0 -11
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-lease-list.test.py +0 -9
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-multihub.test.py +0 -17
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter-output.test.py +0 -34
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/adapter.test.py +0 -39
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/plugin-api-validation.test.py +0 -12
- package/.versions/0.0.4/hermes/plugins/agent-fleet-monitor/dashboard/review-live-wiring.red.test.py +0 -22
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-cap.test.py +0 -15
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-discovery.test.py +0 -19
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-envelope.test.py +0 -13
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-forged-owner.test.py +0 -11
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-lease-list.test.py +0 -9
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-multihub.test.py +0 -17
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter-output.test.py +0 -34
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/adapter.test.py +0 -39
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/plugin-api-validation.test.py +0 -12
- package/.versions/0.0.5/hermes/plugins/agent-fleet-monitor/dashboard/review-live-wiring.red.test.py +0 -22
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-cap.test.py +0 -15
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-discovery.test.py +0 -19
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-envelope.test.py +0 -13
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-forged-owner.test.py +0 -11
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-lease-list.test.py +0 -9
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-multihub.test.py +0 -17
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter-output.test.py +0 -34
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/adapter.test.py +0 -39
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/plugin-api-validation.test.py +0 -12
- package/.versions/0.0.6/hermes/plugins/agent-fleet-monitor/dashboard/review-live-wiring.red.test.py +0 -22
|
@@ -0,0 +1,191 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: using-agent-skills
|
|
3
|
+
description: Discovers and invokes agent skills. Use when starting a session or when you need to discover which skill applies to the current task. This is the meta-skill that governs how all other skills are discovered and invoked.
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Using Agent Skills
|
|
7
|
+
|
|
8
|
+
## Overview
|
|
9
|
+
|
|
10
|
+
Agent Skills is a collection of engineering workflow skills organized by development phase. Each skill encodes a specific process that senior engineers follow. This meta-skill helps you discover and apply the right skill for your current task.
|
|
11
|
+
|
|
12
|
+
## Skill Discovery
|
|
13
|
+
|
|
14
|
+
When a task arrives, identify the development phase and apply the corresponding skill:
|
|
15
|
+
|
|
16
|
+
```
|
|
17
|
+
Task arrives
|
|
18
|
+
│
|
|
19
|
+
├── Don't know what you want yet? ──────→ interview-me
|
|
20
|
+
├── Have a rough concept, need variants? → idea-refine
|
|
21
|
+
├── New project/feature/change? ──→ spec-driven-development
|
|
22
|
+
├── Have a spec, need tasks? ──────→ planning-and-task-breakdown
|
|
23
|
+
├── Implementing code? ────────────→ incremental-implementation
|
|
24
|
+
│ ├── UI work? ─────────────────→ frontend-ui-engineering
|
|
25
|
+
│ ├── API work? ────────────────→ api-and-interface-design
|
|
26
|
+
│ ├── Need better context? ─────→ context-engineering
|
|
27
|
+
│ ├── Need doc-verified code? ───→ source-driven-development
|
|
28
|
+
│ └── Stakes high / unfamiliar code? ──→ doubt-driven-development
|
|
29
|
+
├── Writing/running tests? ────────→ test-driven-development
|
|
30
|
+
│ └── Browser-based? ───────────→ browser-testing-with-devtools
|
|
31
|
+
├── Something broke? ──────────────→ debugging-and-error-recovery
|
|
32
|
+
├── Reviewing code? ───────────────→ code-review-and-quality
|
|
33
|
+
│ ├── Too complex? ─────────────→ code-simplification
|
|
34
|
+
│ ├── Security concerns? ───────→ security-and-hardening
|
|
35
|
+
│ └── Performance concerns? ────→ performance-optimization
|
|
36
|
+
├── Committing/branching? ─────────→ git-workflow-and-versioning
|
|
37
|
+
├── CI/CD pipeline work? ──────────→ ci-cd-and-automation
|
|
38
|
+
├── Deprecating/migrating? ────────→ deprecation-and-migration
|
|
39
|
+
├── Writing docs/ADRs? ───────────→ documentation-and-adrs
|
|
40
|
+
├── Adding logs/metrics/alerts? ───→ observability-and-instrumentation
|
|
41
|
+
└── Deploying/launching? ─────────→ shipping-and-launch
|
|
42
|
+
```
|
|
43
|
+
|
|
44
|
+
## Core Operating Behaviors
|
|
45
|
+
|
|
46
|
+
These behaviors apply at all times, across all skills. They are non-negotiable.
|
|
47
|
+
|
|
48
|
+
### 1. Surface Assumptions
|
|
49
|
+
|
|
50
|
+
Before implementing anything non-trivial, explicitly state your assumptions:
|
|
51
|
+
|
|
52
|
+
```
|
|
53
|
+
ASSUMPTIONS I'M MAKING:
|
|
54
|
+
1. [assumption about requirements]
|
|
55
|
+
2. [assumption about architecture]
|
|
56
|
+
3. [assumption about scope]
|
|
57
|
+
→ Correct me now or I'll proceed with these.
|
|
58
|
+
```
|
|
59
|
+
|
|
60
|
+
Don't silently fill in ambiguous requirements. The most common failure mode is making wrong assumptions and running with them unchecked. Surface uncertainty early — it's cheaper than rework.
|
|
61
|
+
|
|
62
|
+
### 2. Manage Confusion Actively
|
|
63
|
+
|
|
64
|
+
When you encounter inconsistencies, conflicting requirements, or unclear specifications:
|
|
65
|
+
|
|
66
|
+
1. **STOP.** Do not proceed with a guess.
|
|
67
|
+
2. Name the specific confusion.
|
|
68
|
+
3. Present the tradeoff or ask the clarifying question.
|
|
69
|
+
4. Wait for resolution before continuing.
|
|
70
|
+
|
|
71
|
+
**Bad:** Silently picking one interpretation and hoping it's right.
|
|
72
|
+
**Good:** "I see X in the spec but Y in the existing code. Which takes precedence?"
|
|
73
|
+
|
|
74
|
+
### 3. Push Back When Warranted
|
|
75
|
+
|
|
76
|
+
You are not a yes-machine. When an approach has clear problems:
|
|
77
|
+
|
|
78
|
+
- Point out the issue directly
|
|
79
|
+
- Explain the concrete downside (quantify when possible — "this adds ~200ms latency" not "this might be slower")
|
|
80
|
+
- Propose an alternative
|
|
81
|
+
- Accept the human's decision if they override with full information
|
|
82
|
+
|
|
83
|
+
Sycophancy is a failure mode. "Of course!" followed by implementing a bad idea helps no one. Honest technical disagreement is more valuable than false agreement.
|
|
84
|
+
|
|
85
|
+
### 4. Enforce Simplicity
|
|
86
|
+
|
|
87
|
+
Your natural tendency is to overcomplicate. Actively resist it.
|
|
88
|
+
|
|
89
|
+
Before finishing any implementation, ask:
|
|
90
|
+
- Can this be done in fewer lines?
|
|
91
|
+
- Are these abstractions earning their complexity?
|
|
92
|
+
- Would a staff engineer look at this and say "why didn't you just..."?
|
|
93
|
+
|
|
94
|
+
If you build 1000 lines and 100 would suffice, you have failed. Prefer the boring, obvious solution. Cleverness is expensive.
|
|
95
|
+
|
|
96
|
+
### 5. Maintain Scope Discipline
|
|
97
|
+
|
|
98
|
+
Touch only what you're asked to touch.
|
|
99
|
+
|
|
100
|
+
Do NOT:
|
|
101
|
+
- Remove comments you don't understand
|
|
102
|
+
- "Clean up" code orthogonal to the task
|
|
103
|
+
- Refactor adjacent systems as a side effect
|
|
104
|
+
- Delete code that seems unused without explicit approval
|
|
105
|
+
- Add features not in the spec because they "seem useful"
|
|
106
|
+
|
|
107
|
+
Your job is surgical precision, not unsolicited renovation.
|
|
108
|
+
|
|
109
|
+
### 6. Verify, Don't Assume
|
|
110
|
+
|
|
111
|
+
Every skill includes a verification step. A task is not complete until verification passes. "Seems right" is never sufficient — there must be evidence (passing tests, build output, runtime data).
|
|
112
|
+
|
|
113
|
+
Per-skill verification is the local check. The project-wide bar that applies to *every* change, regardless of which skill is active, is the Definition of Done: tests pass, no regressions, behavior verified at runtime, docs updated. See `references/definition-of-done.md`. It complements each task's acceptance criteria rather than replacing them.
|
|
114
|
+
|
|
115
|
+
## Failure Modes to Avoid
|
|
116
|
+
|
|
117
|
+
These are the subtle errors that look like productivity but create problems:
|
|
118
|
+
|
|
119
|
+
1. Making wrong assumptions without checking
|
|
120
|
+
2. Not managing your own confusion — plowing ahead when lost
|
|
121
|
+
3. Not surfacing inconsistencies you notice
|
|
122
|
+
4. Not presenting tradeoffs on non-obvious decisions
|
|
123
|
+
5. Being sycophantic ("Of course!") to approaches with clear problems
|
|
124
|
+
6. Overcomplicating code and APIs
|
|
125
|
+
7. Modifying code or comments orthogonal to the task
|
|
126
|
+
8. Removing things you don't fully understand
|
|
127
|
+
9. Building without a spec because "it's obvious"
|
|
128
|
+
10. Skipping verification because "it looks right"
|
|
129
|
+
|
|
130
|
+
## Skill Rules
|
|
131
|
+
|
|
132
|
+
1. **Check for an applicable skill before starting work.** Skills encode processes that prevent common mistakes.
|
|
133
|
+
|
|
134
|
+
2. **Skills are workflows, not suggestions.** Follow the steps in order. Don't skip verification steps.
|
|
135
|
+
|
|
136
|
+
3. **Multiple skills can apply.** A feature implementation might involve `idea-refine` → `spec-driven-development` → `planning-and-task-breakdown` → `incremental-implementation` → `test-driven-development` → `code-review-and-quality` → `code-simplification` → `shipping-and-launch` in sequence.
|
|
137
|
+
|
|
138
|
+
4. **When in doubt, start with a spec.** If the task is non-trivial and there's no spec, begin with `spec-driven-development`.
|
|
139
|
+
|
|
140
|
+
## Lifecycle Sequence
|
|
141
|
+
|
|
142
|
+
For a complete feature, the typical skill sequence is:
|
|
143
|
+
|
|
144
|
+
```
|
|
145
|
+
1. interview-me → Extract what the user actually wants
|
|
146
|
+
2. idea-refine → Refine vague ideas
|
|
147
|
+
3. spec-driven-development → Define what we're building
|
|
148
|
+
4. planning-and-task-breakdown → Break into verifiable chunks
|
|
149
|
+
5. context-engineering → Load the right context
|
|
150
|
+
6. source-driven-development → Verify against official docs
|
|
151
|
+
7. incremental-implementation → Build slice by slice
|
|
152
|
+
8. observability-and-instrumentation → Instrument as you build (runs parallel with 7-9, not after)
|
|
153
|
+
9. doubt-driven-development → Cross-examine non-trivial decisions in-flight
|
|
154
|
+
10. test-driven-development → Prove each slice works
|
|
155
|
+
11. code-review-and-quality → Review before merge
|
|
156
|
+
12. code-simplification → Reduce unnecessary complexity while preserving behavior
|
|
157
|
+
13. git-workflow-and-versioning → Clean commit history
|
|
158
|
+
14. documentation-and-adrs → Document decisions
|
|
159
|
+
15. deprecation-and-migration → Retire old systems and move users safely when needed
|
|
160
|
+
16. shipping-and-launch → Deploy safely
|
|
161
|
+
```
|
|
162
|
+
|
|
163
|
+
Not every task needs every skill. A bug fix might only need: `debugging-and-error-recovery` → `test-driven-development` → `code-review-and-quality`.
|
|
164
|
+
|
|
165
|
+
## Quick Reference
|
|
166
|
+
|
|
167
|
+
| Phase | Skill | One-Line Summary |
|
|
168
|
+
|-------|-------|-----------------|
|
|
169
|
+
| Define | interview-me | Surface what the user actually wants before any plan, spec, or code exists |
|
|
170
|
+
| Define | idea-refine | Refine ideas through structured divergent and convergent thinking |
|
|
171
|
+
| Define | spec-driven-development | Requirements and acceptance criteria before code |
|
|
172
|
+
| Plan | planning-and-task-breakdown | Decompose into small, verifiable tasks |
|
|
173
|
+
| Build | incremental-implementation | Thin vertical slices, test each before expanding |
|
|
174
|
+
| Build | source-driven-development | Verify against official docs before implementing |
|
|
175
|
+
| Build | doubt-driven-development | Adversarial fresh-context review of every non-trivial decision |
|
|
176
|
+
| Build | context-engineering | Right context at the right time |
|
|
177
|
+
| Build | frontend-ui-engineering | Production-quality UI with accessibility |
|
|
178
|
+
| Build | api-and-interface-design | Stable interfaces with clear contracts |
|
|
179
|
+
| Verify | test-driven-development | Failing test first, then make it pass |
|
|
180
|
+
| Verify | browser-testing-with-devtools | Chrome DevTools MCP for runtime verification |
|
|
181
|
+
| Verify | debugging-and-error-recovery | Reproduce → localize → fix → guard |
|
|
182
|
+
| Review | code-review-and-quality | Five-axis review with quality gates |
|
|
183
|
+
| Review | code-simplification | Preserve behavior while reducing unnecessary complexity |
|
|
184
|
+
| Review | security-and-hardening | OWASP prevention, input validation, least privilege |
|
|
185
|
+
| Review | performance-optimization | Measure first, optimize only what matters |
|
|
186
|
+
| Ship | git-workflow-and-versioning | Atomic commits, clean history |
|
|
187
|
+
| Ship | ci-cd-and-automation | Automated quality gates on every change |
|
|
188
|
+
| Ship | deprecation-and-migration | Remove old systems and migrate users safely |
|
|
189
|
+
| Ship | documentation-and-adrs | Document the why, not just the what |
|
|
190
|
+
| Ship | observability-and-instrumentation | Structured logs, RED metrics, traces, symptom-based alerts |
|
|
191
|
+
| Ship | shipping-and-launch | Pre-launch checklist, monitoring, rollback plan |
|
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,139 @@
|
|
|
1
1
|
# Agent Fleet changelog
|
|
2
2
|
|
|
3
|
+
## 0.0.7
|
|
4
|
+
|
|
5
|
+
### Patch Changes
|
|
6
|
+
|
|
7
|
+
- 9a5091c: Fallback failed or aborted agent-hub model overrides, including local-model memory-limit failures, to the persona's original model before any text or tool work begins.
|
|
8
|
+
- 72ff5e1: Fix `started_at` on bridged Claude Code peers, which reported an uptime of at most 30 seconds.
|
|
9
|
+
|
|
10
|
+
`scripts/coms-claude-bridge.ts` rebuilt its entire registry record on every 30s keepalive, and the builder set `started_at: nowIso()` — so the field never held the start of anything and the Hermes fleet panel showed every bridged Claude pane as freshly started, forever. This is the third copy of the bug that Phase 1 of the fleet-observability work fixed for the two pi harnesses; the bridge now reuses that same `buildLiveRegistryEntry()` instead of keeping its own. Registration captures `started_at` once; the keepalive carries it forward and moves `heartbeat_at`.
|
|
11
|
+
|
|
12
|
+
- 72ff5e1: Show what a fleet agent is actually doing, not merely that it is doing something.
|
|
13
|
+
|
|
14
|
+
The Hermes panel could say `working`. True, and useless — it is the same word for an agent compiling, an agent stuck in a retry loop, and an agent that has been reading the same file for four minutes. `hermes/plugins/agent-fleet-herdr/dashboard/activity.py` reads the agent's own transcript and turns it into a timeline, so a selected row now reads `working · 3m12s · bash git rev-list…` with the steps that led there underneath. It needs nothing from the hub, herdr or a lease: the transcript is written whether or not anybody is watching, which makes this the only part of the panel that also works for a `detached` session no pane hosts.
|
|
15
|
+
|
|
16
|
+
**The slug narrows, `boot` decides.** A cwd routinely holds a dozen transcripts — every resumed session, every restart — and a pi pane and a Claude Code pane can share one exactly. The match is `coms-log/boot.session_id == registry.session_id`, and a candidate whose boot names a different session is skipped rather than used as a fallback. Picking the wrong transcript is not "no data"; it is confident fiction about what another agent is doing.
|
|
17
|
+
|
|
18
|
+
**Bridged Claude Code peers get a tail too, through a link that had to be created.** `coms-claude-bridge.ts` mints its own coms session id and Claude Code has never heard of it, so no shared identifier existed anywhere on disk. `hooks/coms-stop-hook.mjs` now records `transcript_path` alongside the turn text it already writes — an additive key the bridge ignores — and the reader follows registry → herdr pane → hook record → `~/.claude/projects/*.jsonl`, handling the second dialect (`tool_use`/`input`, PascalCase names, `stop_reason`) behind the same projection. A peer whose hook has never fired says so instead of guessing, and because the chain runs through the pane, a _detached_ Claude Code peer cannot be matched at all — the opposite of the pi case, and stated rather than papered over.
|
|
19
|
+
|
|
20
|
+
**Projection, not forwarding.** A transcript holds everything the agent has ever read. `thinking` blocks, `toolResult` output and `user` turns never travel; a tool's arguments pass a per-tool allowlist (`bash` yields its command, `read` a path, `dispatch_agent` the target), a tool nobody listed yields its name and nothing else, and an argument that is a dict where a string was expected renders as nothing rather than being `str()`d into the payload. The two dialects keep separate maps so `Read` cannot inherit `read`'s fields.
|
|
21
|
+
|
|
22
|
+
`GET …/sessions/{project}/{name}/activity?after=<seq>` is cursor-based on a **byte offset into the transcript** — monotone within a file, no state on either side — with reads capped at 256 KB, parses cached on `(path, size, mtime)`, and both ends of the tail cut so a record we split and a record still being appended are never parsed. It returns **no errors**: no transcript, no registry entry, an unreadable file are all a 200 with `available: false` and a reason, because a session with nothing on disk to read is ordinary and a red box every three seconds is not. The herdr lookup a bridged peer needs runs only for `model: "claude-code"`, so a pi peer never pays for it.
|
|
23
|
+
|
|
24
|
+
In the pane it is the modal rather than the row — opening a session is the request for detail, and a transcript read per row per 3 seconds would turn a status panel into a disk load. The description line degrades a piece at a time: an unreadable timestamp drops the age and keeps the action, and no transcript at all leaves the verdict standing alone with the reason underneath.
|
|
25
|
+
|
|
26
|
+
- 72ff5e1: Push fleet events to the herdr pane over a WebSocket instead of waiting for the next poll, without giving up the poll.
|
|
27
|
+
|
|
28
|
+
`WS /api/plugins/agent-fleet-herdr/events/stream?after=<seq>` serves the same ring buffer `GET /events` already serves — same events, same sequence numbers, same cursor — as a first frame carrying the backlog past `after`, then one frame per batch as it happens, and an empty `keepalive` frame every 20s of silence. This is the plan's Phase 5, and it lands last on purpose: latency was only worth spending on once Phases 1–4 had made something worth pushing.
|
|
29
|
+
|
|
30
|
+
**The poll is the contract; the socket is the accelerator.** `ctx.socket` is a documented no-op on OAuth remotes and gives its caller no close event, so the pane never stops polling — it steps `/events` down from every 5s to every 30s while frames are arriving and back up the moment they stop (`shouldPollEvents`). Nothing has to detect a broken socket, because nothing was ever switched off, and the 30s floor is also what recovers a batch the stream dropped. Both feeds run through one handler and one cursor in the renderer, which is what makes them interchangeable rather than additive: `presentEvents(payload, primed, after)` filters by `seq`, so the same event delivered by both is a wasted frame instead of a duplicate toast, and a frame that overtakes another cannot rewind the cursor.
|
|
31
|
+
|
|
32
|
+
**The route has to authenticate itself, which was not obvious.** Every gateway middleware — the auth gate _and_ the one that 404s a plugin missing from `plugins.enabled` — is registered for the `http` scope, so a WebSocket upgrade reaches a plugin router with nothing checked at all. `_socket_gate()` re-runs the gateway's own `_ws_request_is_allowed` / `_ws_auth_ok`, looked up in `sys.modules` rather than re-imported, plus the enabled check. **Unresolvable means refused**: a Hermes that renames those functions gets no stream and a pane that keeps polling, because the alternative is an upgrade quietly converting this into an unauthenticated event feed.
|
|
33
|
+
|
|
34
|
+
**`EventStream` subscribes before it reads the backlog.** The other order has a hole in it — an event landing between the two would be in neither — and this order can only produce a duplicate, which the cursor filter drops. Its per-socket queue is bounded at 64 batches and drops rather than growing without bound; the batch arrives late via the poll instead of never, which is the second reason the poll stays. `Watcher.subscribe()` fans batches out to live listeners on whatever thread observed the snapshot, under the same rule as the outbound sinks: a listener that raises costs a line on stderr and never stops the watcher.
|
|
35
|
+
|
|
36
|
+
`/capabilities` gains `events_stream: true` — not a source, a version marker: backend routes mount at app construction, so the presence of that key is the honest answer to "has Hermes restarted since the plugin changed".
|
|
37
|
+
|
|
38
|
+
The optional half of the phase — replacing the `herdr agent list` subprocess with a JSON-lines socket client — is dropped on measurement rather than taste: seven runs of that command take 3.1–4.6ms, which at a 3s interval is not something a human can perceive, and the alternative is a second implementation of herdr's protocol in Python.
|
|
39
|
+
|
|
40
|
+
- 72ff5e1: Revive the agent-fleet-monitor and fold its task tree into the herdr pane, so a hub's subagents can be watched and stopped from one panel.
|
|
41
|
+
|
|
42
|
+
The monitor was fully wired into `agent-hub` — twelve call sites publishing every child run, its output and its exit — and had never once started, because `monitorLifecycleConfig()` returns `null` without `AGENT_FLEET_PROFILE_ID` and `AGENT_FLEET_MONITOR_RUNTIME_DIR`, and the only thing that ever set them was a manual `export` copied out of a README. Roughly 95 KB of source behind 40 test files was dead in practice. This is the "revive" exit of the plan's Phase 4 decision, taken because the activity tail cannot do the two things the monitor can: show a subagent's raw stdout, and **cancel** it.
|
|
43
|
+
|
|
44
|
+
**The launchers now decide.** `scripts/lib/monitor-env.ts` resolves both variables — the `dev` profile by default, `$XDG_RUNTIME_DIR/agent-fleet-monitor` created mode 0700 — and `scripts/fleet.ts` merges them into every `just fleet` mode. `scripts/team-up.ts` passes them as **pane env** instead, because the hub pane is spawned by a herdr daemon that inherits nothing from the launcher, and a `just fleet team` hub is the one most likely to want a task tree. An operator's own export always wins, `AGENT_FLEET_MONITOR=0` opts out, and a runtime directory that cannot be made 0700 leaves the hub unmonitored rather than handing the Python reader a path it will refuse.
|
|
45
|
+
|
|
46
|
+
**One pane, not two.** `hermes/plugins/agent-fleet-herdr/dashboard/tasks.py` serves `GET …/sessions/{project}/{name}/tasks`, and the selected row's modal grows a subagent tree with each live child's stdout under it and a Cancel beside it. The description line gains the count, because a hub whose own transcript is idle while three specialists work is exactly the case the panel used to render as "nothing is happening". The monitor's separate Desktop panel is left uninstalled — folding the tree in is what makes it unnecessary.
|
|
47
|
+
|
|
48
|
+
**The join is `hubPaneId`, an identifier that already existed on both sides.** A monitor child records `env.HERDR_PANE_ID` of the hub that spawned it; the panel already takes a herdr pane snapshot per poll. One id, written by one process, read by both — no cwd guessing and no "newest hub wins". The consequence is stated rather than hidden: correlation runs through the pane, so a `detached` hub shows no tasks, the same limit bridged Claude Code peers have on the activity tail.
|
|
49
|
+
|
|
50
|
+
**Cancel re-derives ownership before it trusts an id.** The renderer names a task it read from `…/tasks`, but a renderer is not a trusted source: the id is looked up again in a fresh snapshot scoped to that agent's pane, and the hub the cancel is addressed to comes from that snapshot, never from the request. Without it the route would stop any task in any hub on the machine for anyone who could guess an id. `tasks.py` reuses the monitor plugin's own `adapter.py` — 0700 root, 0600 discovery and token files, a recomputed socket hash, lease expiry — rather than keeping a second copy of security-critical discovery code.
|
|
51
|
+
|
|
52
|
+
Task fields pass an explicit allowlist (`ownerSessionId`, `ownerLeaseExpiresAt`, `checkoutId`, `workspaceId` and `hubPaneId` all stop at the backend), output is a bounded 2 KB tail per task with at most 8 socket reads per request, and — as everywhere else in this panel — no monitor, no pane, an unreachable socket and a hub that has spawned nothing are four different sentences and zero errors.
|
|
53
|
+
|
|
54
|
+
Two tests that had been silently stranded by earlier refactors are fixed and wired into `npm test`: `monitor-publisher.test.ts` omitted `parentGeneration`, so the parent lookup failed and the assertion covering this very pane correlation had been red on a clean checkout; and `review-live-wiring.red.test.py` was asserting a bounded cancel response against the two-line re-export shim left behind when the socket server moved, proving only that the shim was short.
|
|
55
|
+
|
|
56
|
+
- 72ff5e1: Replace the Hermes fleet pane's ask box with a per-session modal, and make the panel read-only for now.
|
|
57
|
+
|
|
58
|
+
Selecting a row used to fill in a composer at the bottom of the pane: an input, a Send button, and — squeezed into the header line next to it — the only other thing you could do, `Focus pane`. Sending is withdrawn for the moment, so what is left needed somewhere better than the bottom of a 300px column.
|
|
59
|
+
|
|
60
|
+
A selected row now opens a modal. It carries what the row has to truncate — purpose, model, directory, context use, queue depth, uptime, heartbeat age, and which herdr pane hosts it — and below that the actions available on that agent, which for now is exactly `Focus pane`. An action that is currently impossible stays visible and disabled with the reason beside it, because "why can't I focus this one" is the question the panel exists to answer: a `detached` row is a live session that no pane hosts, which is a different statement from an agent that cannot be reached. A session that dies while its modal is open reads `gone` rather than emptying out, since the selection is re-found in each payload rather than remembered.
|
|
61
|
+
|
|
62
|
+
`presentComposer()` is gone, replaced by `presentSessionMenu()` returning a list of actions — adding one is an entry in that list and a door in the renderer, not a new component. `POST /sessions/{project}/{name}/prompt` and the dispatch transcript behind it are untouched and still work; nothing in the pane calls them.
|
|
63
|
+
|
|
64
|
+
- 72ff5e1: Make a running fleet legible in the Hermes panel, and stop `started_at` from lying.
|
|
65
|
+
|
|
66
|
+
Both harnesses rebuilt their coms registry entry on every 30-second heartbeat with `started_at: nowIso()`, so the field never held the start of anything — a session running for two hours reported an age of a few seconds, and anything trying to show uptime was showing noise. The entry is now built by `buildLiveRegistryEntry()` in `.pi/harnesses/lib/coms-registry-entry.ts`, shared by the standalone coms harness and the copy embedded in agent-hub: registration sets `started_at`, the heartbeat carries it forward and moves `heartbeat_at` instead.
|
|
67
|
+
|
|
68
|
+
The `agent-fleet-herdr` panel forwards what it already had and derives what the renderer cannot. Rows now carry `uptime_s`, `heartbeat_age_s` and `stale` (the same 90s freshness rule the registry reader applies), plus `heartbeat_at` and the pane's `workspace_id`; the payload carries `herdr_panes`, the total pane count from herdr. All time arithmetic happens in the backend, which knows when the snapshot was collected — a `null` timestamp renders as nothing rather than `0s`.
|
|
69
|
+
|
|
70
|
+
That pane count is what finally explains `detached`: "herdr reports no panes at all" and "none of 7 herdr panes reports it" are different problems that used to be the same word. A stopped heartbeat outranks both, because it is a statement about the process rather than about our view of it. Rows also show context use, a non-empty queue and uptime, and a pending dispatch counts up on its own second-by-second instead of looking identical from the moment it is sent to the moment it answers.
|
|
71
|
+
|
|
72
|
+
New: `POST /sessions/{project}/{name}/focus` brings the workspace hosting a peer to the front. It takes `(project, name)` like the prompt endpoint, resolves the workspace server-side from a herdr answer taken now, and validates the id before it reaches argv. A `detached` peer answers 422 — it is still perfectly askable, since coms reaches its own socket, but there is no pane to bring forward.
|
|
73
|
+
|
|
74
|
+
- 72ff5e1: Make the fleet report its own transitions instead of being watched.
|
|
75
|
+
|
|
76
|
+
The Hermes panel could tell you _that_ a session existed and, eventually, _that_ a prompt was answered. Anything in between you had to catch by looking. `hermes/plugins/agent-fleet-herdr/dashboard/watch.py` turns consecutive `/sessions` payloads into events — `needs_answer`, `unblocked`, `finished`, `vanished`, `stale`, `dispatch_answered`, `dispatch_failed` — and hands them to two sinks: a ring buffer the pane drains into toasts, and, only when a config file says so, `hermes send`.
|
|
77
|
+
|
|
78
|
+
It is three layers on purpose. `diff_snapshots(prev, next)` is pure: no I/O, no clock of its own — time comes from the payload's `collected_at` — so every rule is testable from two hand-written fixtures. `Watcher` is the memory around it: a question must persist 20 seconds before it interrupts anyone, an identical event collapses for a minute, and past twelve events in a rolling minute a single `throttled` line says how many were dropped. Only `collect_snapshot` and the runner touch anything.
|
|
79
|
+
|
|
80
|
+
Two rules earn their weight. A herdr outage is not fleet news: when herdr stops answering every row degrades to `unknown`, so that snapshot is discarded whole and the previous one kept — the fleet is not reported as having changed once going blind and once coming back, and whatever really happened is reported on recovery, against evidence. And a question answered inside the debounce suppresses _both_ its `needs_answer` and the `unblocked` that would have followed; announcing the end of something nobody was told about is exactly the noise this layer exists to remove.
|
|
81
|
+
|
|
82
|
+
`GET /events?after=<seq>` is cursor-based, bounded to 200 and per gateway process, and returns `seq` even when the list is empty so a client that fell behind resumes from the present rather than replaying a truncated past. Every `/sessions` request feeds the watcher, and the first one starts a background thread that keeps snapshotting every 15s so closing the pane does not stop the watching; `python3 watch.py --daemon` runs it without a Desktop window, and `--snapshot` prints what the watcher sees.
|
|
83
|
+
|
|
84
|
+
The Telegram sink ships default-off and does not exist unless `$HERMES_HOME/agent-fleet-watch.json` sets `telegram.enabled: true` with a target that survives validation. Target and profile are checked against a character class before they reach argv, `hermes send` is spawned as a list rather than a shell line, and sends run on their own thread behind a bounded queue so a slow subprocess can never hold up a `/sessions` request.
|
|
85
|
+
|
|
86
|
+
In the pane, `needs_answer` is a sticky warning, `vanished` an error and the rest ambient; the toast id is `(kind, project, name)`, so an agent that flaps replaces its toast instead of stacking a column. The events poll deliberately does not pause on `visibilitychange` the way the list poll does — a hidden window is when a toast is worth raising — and its first answer only sets the cursor, so opening the pane never replays history as if it were live.
|
|
87
|
+
|
|
88
|
+
- 72ff5e1: Add the `agent-fleet-herdr` Hermes Desktop plugin: a read-only panel listing live Agent Fleet sessions grouped by coms project, joined to herdr pane state by peer name, surfacing the agents that are waiting for a human answer. Ships both halves (Electron pane + FastAPI backend), a shared `scripts/install-hermes-plugin.sh` installer, and `docs/hermes-desktop-plugins.md` describing the plugin contract and its failure modes.
|
|
89
|
+
- 72ff5e1: Fix herdr presence against herdr 0.7.4+ and let the Agent Fleet panel dispatch prompts.
|
|
90
|
+
|
|
91
|
+
herdr 0.7.4 removed `custom_status` from `pane.report_metadata`, so every presence report had been failing silently — `HerdrPresence.report()` swallows errors — leaving every pane unannotated and every fleet view stuck on `detached`/`unknown`. Presence now writes herdr's `tokens` (`coms`, `proj`, `ctx`, `q`), negotiating the dialect by trying and latching rather than version sniffing, with an `onError` hook that both harnesses log to `coms-log` as `presence_dialect_rejected`.
|
|
92
|
+
|
|
93
|
+
The Claude Code bridge (`scripts/coms-claude-bridge.ts`) had its own second copy of the annotation call and stayed on the removed `custom_status` field, so every bridged Claude peer read `detached` no matter what it was doing. It now goes through the shared `HerdrPresence.annotate()` — annotation only, never `pane.report_agent`, since the bridge polls that same `agent_status` back as its turn-completion signal.
|
|
94
|
+
|
|
95
|
+
Because tokens carry the coms project, the `agent-fleet-herdr` join key is now `(project, name)` instead of `name` — two projects each running an `orchestrator` resolve to their own panes instead of both collapsing to `unknown`. A missing pane is reported as `detached`; `unknown` is now reserved for genuinely competing evidence.
|
|
96
|
+
|
|
97
|
+
The panel gained an ask box: `POST /sessions/{project}/{name}/prompt` hands a coms prompt envelope to a live peer and shows the answer when it lands. The renderer names `(project, name)` only — the socket path is resolved server-side from a freshly re-read registry, so no endpoint reaches a file running with the app's full privileges.
|
|
98
|
+
|
|
99
|
+
- 72ff5e1: Put the Hermes Desktop fleet panel on the front page, and say plainly which subscriptions the fleet runs on.
|
|
100
|
+
|
|
101
|
+
`agent-fleet-herdr` had a thorough runbook and no way in. It is now a first-page section of the README with its own install how-to, a `mermaid` diagram of what it reads, and screenshots of the panel and the session modal — plus the same install path in `docs/getting-started.md`, a two-integration table at the top of `hermes/README.md` (the Desktop panel reads the fleet; the question bridge lets the fleet ask you — easy to conflate, independent to run), and a runtime-layer row in `docs/ARCHITECTURE.md`.
|
|
102
|
+
|
|
103
|
+
`docs/hermes-desktop-plugins.md` gains the part the runbook never had: **how it connects to Agent Fleet**. Four sources, each named with what it contributes and which module writes it — the coms registry (who exists, and the filter), herdr pane presence (what state), the agent's own transcript (what it is doing), the agent-hub monitor (which subagents are running) — and the two consequences that matter to an operator: nothing about launching a fleet changes, and the only write doors are `focus` and subagent `cancel`, both re-derived server-side. The Install section now leads with a prerequisites table, because "an empty panel" is the shared symptom of a missing prerequisite and of a correctly idle fleet.
|
|
104
|
+
|
|
105
|
+
**Bring your own subscriptions — and your own GPUs** is a new README section. Mixing providers inside one fleet is the normal configuration here, not an edge case, and the README never said so: Codex/ChatGPT and GitHub Copilot subscriptions, a real Claude Code pane bridged in as a coms peer, Ollama cloud or local, and locally hosted weights for the always-on cheap roles. It is also the argument for cross-model review — a `builder` and a `code-reviewer` on different labs' models catch what one model rationalizes past — and for the three-tier ladder, where recon runs on your own hardware and only synthesis spends the expensive tier.
|
|
106
|
+
|
|
107
|
+
- 71f70af: Add an optional, profile-aware, fail-closed Hermes watchdog source and lifecycle tooling. Origin-chat delivery, steering, and surgical recovery remain disabled until genuine live Hermes capability evidence exists.
|
|
108
|
+
|
|
109
|
+
The package also carries the backend and Desktop monitor plugin source as opt-in runtime source; nothing is installed, enabled, or launched by installing the package. `agent-fleet set-hermes-watchdog` gains receipt-based adoption of an identical unmanaged skill tree, and the local monitor contract adds `events` and `invoke` alongside the existing snapshot/output/cancel baseline, which is unchanged.
|
|
110
|
+
|
|
111
|
+
Two local transport fixes: the monitor socket no longer drops a response that settles after a half-closing client's FIN, and the watchdog's long-poll read timeout now covers the wait window it requests instead of expiring early and journaling a false outage.
|
|
112
|
+
|
|
113
|
+
Local runtime coverage exercises a real foreground watcher against a disposable Hub socket in observe mode. That evidence is synthetic-local: it proves neither Gate O, live origin delivery, steering, surgical execution, nor A6.
|
|
114
|
+
|
|
115
|
+
- 589d89f: Harden agent-hub against harness-level failures that presented as specialist failures.
|
|
116
|
+
|
|
117
|
+
Session files: an unusable `<agentKey>.json` is now validated and quarantined to `<file>.corrupt-<ISO>` before any spawn, and a run that pi rejects for a session reason is retried exactly once from a clean session. The check is unconditional, because `--session` reaches pi on every run — so neither `team_adjust drop`/`add` nor `/af-agents-restart` could previously recover a corrupt file. Roster messages no longer promise a reusable session file that pi would refuse.
|
|
118
|
+
|
|
119
|
+
Drift watchdog: the session's own `artifacts/`, `findings/`, and `delegations/` subtrees are implicitly in scope, since the deliverable protocol orders specialists to write there, and the judge prompt now says so. The `scope` rule is non-terminal — it reports a drift advisory on the dispatch result instead of stopping the run, matching the post-run scope gate, which reverts nothing. `loop`, `failures`, and `toolcap` stay terminal.
|
|
120
|
+
|
|
121
|
+
Turn budget: input artifacts resolve before a dispatch is counted, and `spawn_research` validates its persona and artifacts before spending a research slot — a path typo no longer costs a budget slot. Artifact paths now resolve across artifact kinds when the basename is unique (the hub writes auto-returns under `returns/`, so `reviews/<x>-run1.md` used to fail), with the correction reported back to the dispatcher and an explicit refusal when two kinds match.
|
|
122
|
+
|
|
123
|
+
Return contract: `ASSERTION A1: PASS — <evidence>` and the bare `A1: PASS` line forms are parsed, with a declared block always winning and an id the schema already classified never re-stated from a loose line. When nothing parses and assertions were tracked, one bounded read-only pass extracts the block from the report already on disk rather than declaring every assertion unproven; extracted returns are labelled as weaker evidence.
|
|
124
|
+
|
|
125
|
+
coms bridge: a Claude Code pane that is mid-turn is now waited on with bounded backoff instead of failing immediately, and `coms_send` accepts `reply_timeout_ms`, which the bridge honours (clamped to one hour) instead of always applying its own default. That default rises from 10 to 30 minutes, matching `coms_await`. Wait-budget exhaustion returns `pending` rather than an error, writes no failure/return artifact, and preserves the original correlation so a late reply can still be collected without re-sending the task.
|
|
126
|
+
|
|
127
|
+
Assertion ledger: every assertion must name its `source` (the plan line, user request, or finding it encodes) — a sourceless batch is refused by id, because a specialist told to prove `A9` previously had to spend a dispatch and an ASK_USER cycle asking where `A9` came from. Batches over 8 open assertions are accepted with a warning suggesting the split. The orchestrator persona documents both rules.
|
|
128
|
+
|
|
129
|
+
Context accounting: each specialist and research helper is now measured against **its own** model's context window, resolved from pi's model registry with the source recorded, instead of against the dispatcher's window — which is why readings like "Planner context at 315%" were unactionable. A reading above 100% emits a one-time diagnostic naming the window and its source, a session at or past a full window is recycled unconditionally, and a resumed session whose projected prompt would overflow is recycled **before** the spawn rather than after 985s of billed work.
|
|
130
|
+
|
|
131
|
+
Concurrency: requests to one provider are capped per process (default 2 in flight for `custom/*`, unlimited elsewhere, configurable via `AGENT_HUB_PROVIDER_LIMITS`, e.g. `custom=4` or `custom=off`). Queued children still run; a nested spawn reuses its parent's permit so it can never wait on its own ancestor. Read-only children (and research helpers) now retry once on their fallback model when the provider fails **mid-run** — previously only a pre-work failure was recoverable, so a local-endpoint OOM discarded the whole run. Write-capable children keep the strict rule, since a retry there could duplicate edits.
|
|
132
|
+
|
|
133
|
+
Peer visibility: `coms_list` reports each peer's `pane_id` and `status` (`idle` | `working` | `booting`), so a sender can check addressability instead of screen-scraping the pane — the gap behind 127 `herdr_read_pane` calls in one session. `herdr_spawn_peer` waits (bounded) for the spawned peer to register and returns `peer_ready` with its coms name instead of only a pane id, and peers that were spawned but never sent to are named at turn end and in `/af-hub-report`; closing still requires the human's confirmation.
|
|
134
|
+
|
|
135
|
+
Delivery failures: a run that errored or timed out writes its output to `artifacts/failures/<agentKey>-run<N>.md`, never `returns/`, and the dispatch result names it a delivery failure with no assertion evidence. A 142-byte coms error stub stored as a return previously cost a full dispatch investigating a review that had in fact succeeded — only its reply was lost.
|
|
136
|
+
|
|
3
137
|
## 0.0.6
|
|
4
138
|
|
|
5
139
|
### Patch Changes
|
package/README.md
CHANGED
|
@@ -3,20 +3,51 @@
|
|
|
3
3
|
[](https://www.npmjs.com/package/@chankov/agent-fleet)
|
|
4
4
|
[](LICENSE)
|
|
5
5
|
[](#quick-start)
|
|
6
|
+
[](#bring-your-own-subscriptions--and-your-own-gpus)
|
|
7
|
+
[](#watch-the-fleet-from-your-desktop--the-hermes-plugin)
|
|
6
8
|
|
|
7
9
|
**Operate a coding-agent *fleet* — not just one chat session.**
|
|
8
10
|
|
|
9
|
-
Agent Fleet is a **multi-agent orchestration system for AI coding agents**, built pi-first: a thin dispatcher routes work to specialist agents under a Verification Contract, whole teams spawn as tiled workspaces you can snapshot and resume, agents message each other over a shared coms plane — and a library of **29 production-grade lifecycle skills** and **15 personas** keeps every agent disciplined.
|
|
11
|
+
Agent Fleet is a **multi-agent orchestration system for AI coding agents**, built pi-first: a thin dispatcher routes work to specialist agents under a Verification Contract, whole teams spawn as tiled workspaces you can snapshot and resume, agents message each other over a shared coms plane, a **Hermes Desktop panel** shows you the whole fleet at a glance — and a library of **29 production-grade lifecycle skills** and **15 personas** keeps every agent disciplined.
|
|
10
12
|
|
|
11
|
-
|
|
13
|
+
Three things make it unusual:
|
|
12
14
|
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
15
|
+
1. **You run it on the subscriptions you already pay for** — Codex/ChatGPT, GitHub Copilot, Claude, Ollama — mixed *inside one fleet*, alongside models running locally on your own GPU. [How ↓](#bring-your-own-subscriptions--and-your-own-gpus)
|
|
16
|
+
2. **The dispatcher never drowns.** Research output, subagent dumps and specialist chatter stay *out* of the orchestrator's context window — on disk, behind file paths. [How ↓](#agent-hub-a-thin-context-dispatcher-for-pi)
|
|
17
|
+
3. **You can actually see it.** Tiled panes in [herdr](https://herdr.dev), a live session panel in **Hermes Desktop**, and questions relayed to your **phone** when an agent needs a human. [How ↓](#watch-the-fleet-from-your-desktop--the-hermes-plugin)
|
|
18
|
+
|
|
19
|
+

|
|
20
|
+
|
|
21
|
+
---
|
|
22
|
+
|
|
23
|
+
## The fleet at a glance
|
|
24
|
+
|
|
25
|
+
```mermaid
|
|
26
|
+
flowchart TD
|
|
27
|
+
You(["<b>You</b><br/>terminal · desktop · phone"])
|
|
28
|
+
|
|
29
|
+
subgraph Control["Control surfaces"]
|
|
30
|
+
Herdr["<b>herdr</b><br/>tiled peer workspaces<br/>snapshot · resume"]
|
|
31
|
+
Desktop["<b>Hermes Desktop plugin</b><br/>live session panel<br/>who is blocked · what it is doing"]
|
|
32
|
+
Phone["<b>Hermes relay</b><br/>hub questions → Telegram<br/>answer from anywhere"]
|
|
33
|
+
end
|
|
34
|
+
|
|
35
|
+
Hub["<b>agent-hub</b> — thin dispatcher<br/>owns the Verification Contract on disk"]
|
|
36
|
+
|
|
37
|
+
subgraph Team["Specialist team (.pi/agents/teams.yaml)"]
|
|
38
|
+
Specialists["planner · plan-reviewer · builder<br/>test-engineer · code-reviewer · documenter"]
|
|
39
|
+
end
|
|
40
|
+
|
|
41
|
+
Research["<b>research helpers</b> — read-only<br/>findings go to disk, never to the dispatcher"]
|
|
42
|
+
Coms["<b>coms</b> — peer data plane<br/>pi ⇄ pi ⇄ Claude Code panes"]
|
|
43
|
+
|
|
44
|
+
You --> Control
|
|
45
|
+
Control --> Hub
|
|
46
|
+
Hub --> Specialists
|
|
47
|
+
Specialists --> Research
|
|
48
|
+
Hub <--> Coms
|
|
49
|
+
Specialists <--> Coms
|
|
50
|
+
Coms --> Desktop
|
|
20
51
|
```
|
|
21
52
|
|
|
22
53
|
| Layer | Job |
|
|
@@ -24,7 +55,8 @@ Agent Fleet is a **multi-agent orchestration system for AI coding agents**, buil
|
|
|
24
55
|
| **[agent-hub](#agent-hub-a-thin-context-dispatcher-for-pi)** | Thin-context dispatcher on **pi** — drives specialists under a Verification Contract |
|
|
25
56
|
| **[herdr](https://herdr.dev)** | Fleet control plane — tiled peer workspaces, presence, snapshot/resume |
|
|
26
57
|
| **coms** | Peer data plane — bidirectional messaging between agents (including Claude Code panes) |
|
|
27
|
-
| **Hermes** |
|
|
58
|
+
| **[Hermes Desktop plugin](#watch-the-fleet-from-your-desktop--the-hermes-plugin)** | Live fleet panel in the Hermes desktop app — every session, its state, what it is doing, and who is waiting on you |
|
|
59
|
+
| **[Hermes relay](docs/coms-hermes-bridge.md)** | Remote human control — relay hub questions to your phone and answer them there |
|
|
28
60
|
| **Codex conductor (experimental)** | Android-initiated, approval-gated delegation to live coms peers through a user-systemd service |
|
|
29
61
|
| **[Skills](docs/skills-catalog.md) + [personas](docs/agents.md)** | Lifecycle discipline — how to spec, plan, build, verify, review, and ship |
|
|
30
62
|
|
|
@@ -90,7 +122,131 @@ Updates flow through `git pull`. Symlinks need Developer Mode on Windows.
|
|
|
90
122
|
|
|
91
123
|
</details>
|
|
92
124
|
|
|
93
|
-
|
|
125
|
+
Versioned with [semver](https://semver.org) — [CHANGELOG.md](CHANGELOG.md) · [docs/npm-install.md](docs/npm-install.md).
|
|
126
|
+
|
|
127
|
+
---
|
|
128
|
+
|
|
129
|
+
## Bring your own subscriptions — and your own GPUs
|
|
130
|
+
|
|
131
|
+
**There is no Agent Fleet API key.** Every agent in the fleet declares which model it wants, and pi resolves that against whatever providers *you* are already signed in to. A single fleet routinely mixes several at once — that is the normal configuration, not an edge case.
|
|
132
|
+
|
|
133
|
+
| Where the tokens come from | Looks like | Who uses it in a fleet |
|
|
134
|
+
|---|---|---|
|
|
135
|
+
| **ChatGPT / Codex subscription** | `openai-codex/gpt-5.6-terra`, `…-sol`, `…-luna`, `…-codex-spark` | Default for most shipped personas |
|
|
136
|
+
| **GitHub Copilot subscription** | `github-copilot/claude-sonnet-4.6`, `github-copilot/claude-haiku-4.5` | Common override for `builder` / reviewers |
|
|
137
|
+
| **Claude subscription** | a real **Claude Code pane**, bridged in as a first-class coms peer | `plan-reviewer` and `code-reviewer` — cross-model review |
|
|
138
|
+
| **Ollama — cloud *or* local** | `ollama/glm-5.2:cloud`, `ollama/minimax-m3:cloud`, or a model on your own box | Overflow capacity and cost control |
|
|
139
|
+
| **Anything else pi can address** | local MLX / llama.cpp / LM Studio weights, e.g. a 4-bit `Qwen3.6-35B-A3B` orchestrator | Cheap always-on roles: dispatcher, recon, docs |
|
|
140
|
+
|
|
141
|
+
Two consequences worth spelling out:
|
|
142
|
+
|
|
143
|
+
- **Cross-model review is the point, not a party trick.** A `builder` on one lab's model and a `code-reviewer` on another's catches what a single model rationalizes past. `plan-reviewer` and `code-reviewer` ship ready to run as **Claude Code peers** for exactly this.
|
|
144
|
+
- **Cost lives on a ladder, not a switch.** Each persona declares a default plus a switch list over a three-tier policy — *deep reasoning / workhorse / fast recon* — swappable at runtime per persona (`/af-agent-model`) or fleet-wide (`/af-models <profile>`). Recon sweeps run on the cheap tier or on your own hardware; only synthesis and verdicts spend the expensive one.
|
|
145
|
+
|
|
146
|
+
```bash
|
|
147
|
+
/af-models fast # move the whole team down a tier
|
|
148
|
+
/af-agent-model builder github-copilot/claude-sonnet-4.6
|
|
149
|
+
/af-agent-models-substitute openai-codex/gpt-5.6-sol ollama/glm-5.2:cloud # swap one model everywhere
|
|
150
|
+
```
|
|
151
|
+
|
|
152
|
+
Per-project defaults live in `.ai/agent-fleet-overrides.md` — see [docs/agent-fleet-setup.md](docs/agent-fleet-setup.md).
|
|
153
|
+
|
|
154
|
+
---
|
|
155
|
+
|
|
156
|
+
## Watch the fleet from your desktop — the Hermes plugin
|
|
157
|
+
|
|
158
|
+
Once a fleet is more than two agents, "is anything waiting on me?" stops being answerable by looking at panes. **`agent-fleet-herdr`** is an in-repo **Hermes Desktop** plugin that answers it: a live panel of every Agent Fleet session, grouped by project, sorted so that **the agent blocked on a human is at the top**.
|
|
159
|
+
|
|
160
|
+

|
|
161
|
+
|
|
162
|
+
**What the panel gives you**
|
|
163
|
+
|
|
164
|
+
- **Every session, not every pane.** The list is built from the coms registry, so an agent that left its herdr pane still shows up — as `detached`, which is a different fact from "gone".
|
|
165
|
+
- **A toast when something needs you.** `needs_answer` (an agent blocked on a human for more than 20s), `vanished` (a session that died mid-work), `stale`, `finished` — pushed over a WebSocket, with a poll underneath that never switches off. Optional [Telegram fan-out](docs/hermes-desktop-plugins.md#being-told-instead-of-watching) for alerts that must survive a closed window.
|
|
166
|
+
- **What an agent is doing *right now*.** Select a row and the modal reads that agent's own transcript — `bash git rev-list…`, `dispatch_agent builder`, `update_assertion` — projected through a per-tool allowlist, never forwarded wholesale.
|
|
167
|
+
- **The subagents underneath it.** A hub whose own transcript is idle while three specialists work is exactly the case a naive status panel renders as "nothing happening". Each live child carries a tail of its stdout and a **Cancel** button.
|
|
168
|
+
- **One action, and honest disabled states.** `Focus pane` brings the workspace hosting that agent to the front. An action that cannot work right now stays visible with its reason beside it.
|
|
169
|
+
|
|
170
|
+

|
|
171
|
+
|
|
172
|
+
### How it connects to Agent Fleet
|
|
173
|
+
|
|
174
|
+
The plugin reads what the fleet already writes. It adds no daemon to your agents and holds no write door into them.
|
|
175
|
+
|
|
176
|
+
```mermaid
|
|
177
|
+
flowchart LR
|
|
178
|
+
subgraph Fleet["Your running fleet"]
|
|
179
|
+
Agents["pi + Claude Code peers"]
|
|
180
|
+
Registry[("~/.pi/coms/projects/*<br/><b>coms registry</b><br/>who exists")]
|
|
181
|
+
Transcripts[("~/.pi/agent/sessions<br/>~/.claude/projects<br/><b>transcripts</b><br/>what it is doing")]
|
|
182
|
+
HerdrD["<b>herdr</b><br/>what state it is in"]
|
|
183
|
+
Agents --> Registry
|
|
184
|
+
Agents --> Transcripts
|
|
185
|
+
Agents --> HerdrD
|
|
186
|
+
end
|
|
187
|
+
|
|
188
|
+
subgraph Hermes["Hermes"]
|
|
189
|
+
API["<b>backend</b> — FastAPI<br/>plugins/agent-fleet-herdr/dashboard<br/>join · watch · project"]
|
|
190
|
+
Pane["<b>Desktop pane</b> — ESM<br/>desktop-plugins/agent-fleet-herdr"]
|
|
191
|
+
API -->|"REST + WS"| Pane
|
|
192
|
+
end
|
|
193
|
+
|
|
194
|
+
Registry --> API
|
|
195
|
+
Transcripts --> API
|
|
196
|
+
HerdrD --> API
|
|
197
|
+
Pane -->|"POST focus"| HerdrD
|
|
198
|
+
```
|
|
199
|
+
|
|
200
|
+
- **The coms registry is the filter and the source of truth for *who exists*.** A herdr pane with no registry entry is not an Agent Fleet session and is not shown.
|
|
201
|
+
- **herdr says what state each one is in**, joined by the `(project, name)` pair a pane advertises — not by cwd, which a pi pane and a Claude Code pane in the same repo would collide on.
|
|
202
|
+
- **The agent's own transcript says what it is doing.** That file is written whether or not anybody is watching, which is why the activity tail also works for a `detached` session no pane hosts.
|
|
203
|
+
- **Reads only, except two doors.** `focus` and subagent `cancel` — both re-derived server-side from a fresh snapshot, so a renderer cannot name a target it was not given.
|
|
204
|
+
|
|
205
|
+
### Install it
|
|
206
|
+
|
|
207
|
+
**Prerequisites:** Hermes v0.19.0+ (the `hermes` CLI on `PATH`, plus the Desktop app), this repo checked out, and a fleet that has run at least once (so `~/.pi/coms/projects/` exists). [herdr](https://herdr.dev) is optional — without it every row reads `unknown` instead of a live state.
|
|
208
|
+
|
|
209
|
+
```bash
|
|
210
|
+
scripts/install-hermes-plugin.sh agent-fleet-herdr
|
|
211
|
+
```
|
|
212
|
+
|
|
213
|
+
That one command backs up `config.yaml`, symlinks **both halves** into the profile (`plugins/` for the FastAPI backend, `desktop-plugins/` for the pane), runs `hermes plugins enable`, and verifies the result — then **prints** the restart steps instead of performing them, because restarting a live gateway is your call:
|
|
214
|
+
|
|
215
|
+
```bash
|
|
216
|
+
# 1. restart the Hermes Desktop app ← the pane talks to the gateway Desktop spawns for itself
|
|
217
|
+
# 2. hermes gateway restart ← only if the web dashboard / TUI needs the routes too
|
|
218
|
+
```
|
|
219
|
+
|
|
220
|
+
Then start a fleet and open the **Agent Fleet** tab:
|
|
221
|
+
|
|
222
|
+
```bash
|
|
223
|
+
just fleet team default # monitored hub + peers; nothing extra to configure
|
|
224
|
+
```
|
|
225
|
+
|
|
226
|
+
<details>
|
|
227
|
+
<summary><b>Options and verification</b></summary>
|
|
228
|
+
|
|
229
|
+
```bash
|
|
230
|
+
scripts/install-hermes-plugin.sh agent-fleet-herdr --profile dev # a non-default profile
|
|
231
|
+
scripts/install-hermes-plugin.sh agent-fleet-herdr --copy # no symlinks
|
|
232
|
+
scripts/install-hermes-plugin.sh agent-fleet-herdr --dry-run # print, change nothing
|
|
233
|
+
scripts/install-hermes-plugin.sh agent-fleet-herdr --uninstall
|
|
234
|
+
```
|
|
235
|
+
|
|
236
|
+
Working install: `~/.hermes/logs/gui.log` contains `Mounted plugin API routes: /api/plugins/agent-fleet-herdr/` at Desktop start, and `/capabilities` reports both sources. An empty panel with no error means the enable gate 404'd — that is documented, along with every other failure mode that looks identical from the outside, in the runbook.
|
|
237
|
+
|
|
238
|
+
</details>
|
|
239
|
+
|
|
240
|
+
**Deep dive:** [docs/hermes-desktop-plugins.md](docs/hermes-desktop-plugins.md) — the plugin contract, the four rules that cost the most to learn, the full API, and the deliberate limits.
|
|
241
|
+
|
|
242
|
+
### And on your phone
|
|
243
|
+
|
|
244
|
+
Separately from the panel, the **Hermes relay** pipes an agent's `ask_user` question — choices intact, not flattened into a message — to Telegram, and races your phone's answer against a local one. Whichever arrives first wins; the fleet keeps working meanwhile.
|
|
245
|
+
|
|
246
|
+
See [docs/coms-hermes-bridge.md](docs/coms-hermes-bridge.md) and the [screenshots](hermes/README.md#integration-in-action).
|
|
247
|
+
|
|
248
|
+
<details>
|
|
249
|
+
<summary><b>Experimental: delegate to live peers from ChatGPT Android</b></summary>
|
|
94
250
|
|
|
95
251
|
The optional Codex Remote-Control conductor is verified on Linux with Codex CLI `0.144.x`. Hermes remains the inbound `ask_user` route; Codex is outbound-only and delegates one confirmed task at a time to peers already visible in the same coms project.
|
|
96
252
|
|
|
@@ -106,7 +262,7 @@ In ChatGPT Android, open the paired Remote Control host and use the managed exte
|
|
|
106
262
|
|
|
107
263
|
Lifecycle, approval flow, examples, recovery, and security boundaries: **[Codex Remote-Control conductor runbook](docs/codex-remote-conductor.md)**.
|
|
108
264
|
|
|
109
|
-
|
|
265
|
+
</details>
|
|
110
266
|
|
|
111
267
|
---
|
|
112
268
|
|
|
@@ -114,15 +270,15 @@ Versioned with [semver](https://semver.org) — [CHANGELOG.md](CHANGELOG.md) ·
|
|
|
114
270
|
|
|
115
271
|
`agent-hub` turns a single **pi** session into a **dispatcher that drives a live team of specialist subagents** — planner, builder, reviewer, test-engineer, documenter — with read-only research helpers fanning out beneath them, peer-to-peer `coms` messaging embedded, and a `damage-control-continue` guardrail on every tool call.
|
|
116
272
|
|
|
273
|
+

|
|
274
|
+
|
|
117
275
|
What makes it different is what it **doesn't** put in front of the dispatcher LLM. Multi-agent setups usually drown the orchestrator: every subagent's output, every research dump flows back into one context window until it compacts and forgets. `agent-hub` is built the other way around:
|
|
118
276
|
|
|
119
277
|
- **Research never enters the dispatcher context.** Specialists end their turn with `NEEDS_RESEARCH:` lines; the hub fans out read-only helpers, writes findings to disk, and resumes the specialist with file paths. The dispatcher sees a one-line notice — never the raw findings. Each local-disk research tool call has a parent-side 120-second watchdog (configurable as `recon-search-timeout-s` in `.ai/agent-fleet-overrides.md`), not a whole-agent deadline.
|
|
120
278
|
- **The Verification Contract lives on disk.** A ledger of checkable acceptance assertions, built from the request *before* any builder runs, rendered as one status line (`Assertions: 2✓ 1○ 1✗ · open: A4`). A stated requirement is never silently dropped, and the contract survives compaction.
|
|
121
279
|
- **Specialists run `--no-extensions`.** Tools and credentials stay scoped to the subagent that needs them instead of leaking up into the dispatcher.
|
|
122
280
|
|
|
123
|
-

|
|
126
282
|
|
|
127
283
|
```bash
|
|
128
284
|
just fleet hub # guarded dispatcher + research + coms + orchestrator persona
|
|
@@ -135,7 +291,7 @@ just fleet down docs # snapshot + close cleanly
|
|
|
135
291
|
just fleet resume docs # rebuild the grid; pi peers continue their conversations
|
|
136
292
|
```
|
|
137
293
|
|
|
138
|
-
Deep dive: [agent-hub harness README](.pi/harnesses/agent-hub/README.md) (the full dispatch loop, coms layer, configuration) · [fleet hierarchy](docs/ARCHITECTURE.md#fleet-hierarchy) · [pi extension catalog](docs/pi-extensions.md) · [Claude Code coms bridge](docs/claude-code-coms-bridge.md) · [Hermes bridge](docs/coms-hermes-bridge.md)
|
|
294
|
+
Deep dive: [agent-hub harness README](.pi/harnesses/agent-hub/README.md) (the full dispatch loop, coms layer, configuration) · [fleet hierarchy](docs/ARCHITECTURE.md#fleet-hierarchy) · [pi extension catalog](docs/pi-extensions.md) · [Claude Code coms bridge](docs/claude-code-coms-bridge.md) · [Hermes bridge](docs/coms-hermes-bridge.md).
|
|
139
295
|
|
|
140
296
|
---
|
|
141
297
|
|
|
@@ -192,7 +348,7 @@ Full roster, skill hooks, install matrix, and team composition: **[docs/agents.m
|
|
|
192
348
|
|
|
193
349
|
## Why Agent Fleet?
|
|
194
350
|
|
|
195
|
-
One coding agent is an assistant; a *fleet* is a team you operate. Agent Fleet exists for the moment a single session stops being enough — when you want a dispatcher driving specialists under a Verification Contract, whole peer teams you can snapshot and resume, Claude Code panes that answer pi agents mid-task, and a phone-reachable human in the loop.
|
|
351
|
+
One coding agent is an assistant; a *fleet* is a team you operate. Agent Fleet exists for the moment a single session stops being enough — when you want a dispatcher driving specialists under a Verification Contract, whole peer teams you can snapshot and resume, Claude Code panes that answer pi agents mid-task, a desktop panel that tells you which agent is stuck, and a phone-reachable human in the loop.
|
|
196
352
|
|
|
197
353
|
Discipline is the other half. AI coding agents default to the shortest path — skipping specs, tests, and security reviews. The skill library gives every agent in the fleet the same discipline senior engineers bring to production code, baking in practices from [Software Engineering at Google](https://abseil.io/resources/swe-book) and Google's [engineering practices guide](https://google.github.io/eng-practices/): Hyrum's Law in API design, the test pyramid and Beyonce Rule in testing, change sizing in review, Chesterton's Fence in simplification, trunk-based development in git workflow.
|
|
198
354
|
|
|
@@ -210,7 +366,8 @@ Discipline is the other half. AI coding agents default to the shortest path —
|
|
|
210
366
|
| [docs/agents.md](docs/agents.md) | All 15 personas: roster, skill hooks, install matrix, teams |
|
|
211
367
|
| [docs/pi-setup.md](docs/pi-setup.md) · [docs/pi-extensions.md](docs/pi-extensions.md) | pi install paths, harnesses, and utility extensions |
|
|
212
368
|
| [docs/opencode-setup.md](docs/opencode-setup.md) | OpenCode setup and `af-*` commands |
|
|
213
|
-
| [docs/agent-fleet-setup.md](docs/agent-fleet-setup.md) | Per-project overrides (`.ai/agent-fleet-overrides.md`) — spec/plan paths, dev server, branch policy, dispatcher language, rules/docs targets |
|
|
369
|
+
| [docs/agent-fleet-setup.md](docs/agent-fleet-setup.md) | Per-project overrides (`.ai/agent-fleet-overrides.md`) — spec/plan paths, dev server, branch policy, per-persona models, dispatcher language, rules/docs targets |
|
|
370
|
+
| **[docs/hermes-desktop-plugins.md](docs/hermes-desktop-plugins.md)** | **The Hermes Desktop plugin: install, contract, API, failure modes, limits** |
|
|
214
371
|
| [docs/claude-code-coms-bridge.md](docs/claude-code-coms-bridge.md) · [docs/coms-hermes-bridge.md](docs/coms-hermes-bridge.md) · [docs/codex-remote-conductor.md](docs/codex-remote-conductor.md) | Claude Code as a coms peer · phone relay · experimental Codex remote-control operator runbook |
|
|
215
372
|
| [docs/npm-install.md](docs/npm-install.md) | CLI reference, versioning, update flow |
|
|
216
373
|
| [references/](references/) | 9 checklists skills pull in: testing, security, performance, accessibility, observability, orchestration + fleet-coordination + prompting patterns |
|