fix(metis): switch primary model to claude-sonnet-4-6 + correct AGENTS.md inaccuracies

Source code change:
- src/shared/model-requirements.ts: prepend claude-sonnet-4-6 to metis fallback
  chain so Sonnet becomes the default. Opus 4.7 max remains as the immediate
  fallback for callers who want extra reasoning.
- src/shared/model-requirements.test.ts: update assertion to expect Sonnet
  primary + Opus secondary.

AGENTS.md accuracy fixes (verified against source):
- Agent modes: Sisyphus/Hephaestus are 'primary' (not 'all'); Sisyphus-Junior
  is 'subagent' (not 'all'). Confirmed via 'const MODE: AgentMode = ...' in
  each agent file. Also clarified Prometheus has no agentSources factory and
  is built via buildPrometheusAgentConfig.
- Sisyphus fallback chain: corrected order to kimi-k2.6 → k2p5 → kimi-k2.5
  → gpt-5.5 medium → glm-5 → big-pickle (was missing kimi-k2.5).
- Librarian/Explore: added missing minimax-m2.7 step between -highspeed and
  claude-haiku-4-5.
- Metis chain: removed fictitious gemini-3.1-pro entry.
- Sisyphus-Junior chain: spelled out the actual fallback (was 'user-configurable').
- Temperatures: Sisyphus/Hephaestus do not set explicit temperature (model
  default); Sisyphus-Junior is 0.1 via SISYPHUS_JUNIOR_DEFAULTS.
- Quick category default: gpt-5.4-mini (not gpt-5.4-mini-fast).

Team-mode corrections:
- Eligibility registry has 3 verdicts: eligible (sisyphus, atlas, sisyphus-junior),
  conditional (hephaestus — needs D-36 teammate permission), hard-reject
  (oracle, librarian, explore, multimodal-looker, metis, momus, prometheus).
- Schema has 11 fields, not 4: added max_messages_per_run, max_wall_clock_minutes,
  max_member_turns, base_dir, message_payload_max_bytes, recipient_unread_max_bytes,
  mailbox_poll_interval_ms.
- Hooks: 'team-session-events' is 4 sub-handlers in src/plugin/event.ts
  (team-idle-wake-hint, team-lead-orphan-handler, team-member-error-handler,
  team-member-status-handler), not a single Continuation-tier hook.
- Tier counts now show base + team-mode: ToolGuard 14/15, Transform 5/7.
- Total: 52 base hooks, 59 with team-mode.

Doc cascade for the Metis change:
- docs/guide/orchestration.md, agent-model-matching.md, installation.md
- docs/reference/configuration.md, features.md
This commit is contained in:
YeonGyu-Kim
2026-05-08 13:06:34 +09:00
parent 838b5ae216
commit 2dfa6336f5
29 changed files with 2023 additions and 104 deletions
+26 -18
View File
@@ -28,8 +28,8 @@ serverPlugin(input, options)
3. detectExternalSkillPlugin() # warn if conflicting plugin loaded
4. injectServerAuthIntoClient() # wire auth headers into shared SDK client
5. loadPluginConfig() # walk project + user JSONC → Zod safeParse → migrate
6. initializeOpenClaw() # if openclaw config present (start reply-listener daemon)
6. checkTeamModeDependencies() # if team_mode.enabled (verify git, tmux, ensure ~/.omo/teams/)
6a. initializeOpenClaw() # if openclaw config present (start reply-listener daemon)
6b. checkTeamModeDependencies() # if team_mode.enabled (verify git, tmux, ensure ~/.omo/teams/)
7. createManagers/Tools/Hooks/PluginInterface
```
@@ -50,31 +50,39 @@ loadPluginConfig(directory, ctx)
## HOOK COMPOSITION (5-tier)
Counts verified from each composer's return object. Numbers in brackets show counts when `team_mode.enabled`.
```
createHooks()
├─→ createCoreHooks()
│ ├─ createSessionHooks() # 24: contextWindowMonitor, thinkMode, ralphLoop, modelFallback,
│ │ runtimeFallback, anthropicEffort, anthropicContextWindowLimitRecovery,
│ │ autoUpdateChecker, agentUsageReminder, nonInteractiveEnv,
│ │ interactiveBashSession, editErrorRecovery, delegateTaskRetry,
│ │ startWork, prometheusMdOnly, sisyphusJuniorNotepad,
│ │ questionLabelTruncator, taskResumeInfo, noSisyphusGpt,
│ │ noHephaestusNonGpt, legacyPluginToast, sessionRecovery,
│ │ sessionNotification, preemptiveCompaction
│ ├─ createToolGuardHooks() # 14: commentChecker, toolOutputTruncator, directoryAgentsInjector,
│ │ directoryReadmeInjector, emptyTaskResponseDetector, rulesInjector,
│ │ tasksTodowriteDisabler, writeExistingFileGuard, bashFileReadGuard,
│ │ readImageResizer, todoDescriptionOverride, webfetchRedirectGuard,
│ │ hashlineReadEnhancer, jsonErrorRecovery
└─ createTransformHooks() # 5: claudeCodeHooks, keywordDetector, contextInjectorMessagesTransform,
thinkingBlockValidator, toolPairValidator
│ ├─ createSessionHooks() # 24: contextWindowMonitor, preemptiveCompaction, sessionRecovery,
│ │ sessionNotification, thinkMode, modelFallback,
│ │ anthropicContextWindowLimitRecovery, autoUpdateChecker,
│ │ agentUsageReminder, nonInteractiveEnv, interactiveBashSession,
│ │ ralphLoop, editErrorRecovery, delegateTaskRetry, startWork,
│ │ prometheusMdOnly, sisyphusJuniorNotepad, noSisyphusGpt,
│ │ noHephaestusNonGpt, questionLabelTruncator, taskResumeInfo,
│ │ anthropicEffort, runtimeFallback, legacyPluginToast
│ ├─ createToolGuardHooks() # 14 [+1 with team-mode]: commentChecker, toolOutputTruncator,
│ │ directoryAgentsInjector, directoryReadmeInjector,
│ │ emptyTaskResponseDetector, rulesInjector, tasksTodowriteDisabler,
│ │ writeExistingFileGuard, bashFileReadGuard, hashlineReadEnhancer,
│ │ jsonErrorRecovery, readImageResizer, todoDescriptionOverride,
│ webfetchRedirectGuard [+ teamToolGating]
└─ createTransformHooks() # 5 [+2 with team-mode]: claudeCodeHooks, keywordDetector,
│ contextInjectorMessagesTransform, thinkingBlockValidator,
│ toolPairValidator [+ teamModeStatusInjector, teamMailboxInjector]
├─→ createContinuationHooks() # 7: stopContinuationGuard, compactionContextInjector,
│ compactionTodoPreserver, todoContinuationEnforcer (boulder),
│ unstableAgentBabysitter, backgroundNotificationHook, atlasHook
└─→ createSkillHooks() # 2: categorySkillReminder, autoSlashCommand
Direct event handlers (src/plugin/event.ts, when team_mode.enabled): +4
team-idle-wake-hint, team-lead-orphan-handler,
team-member-error-handler, team-member-status-handler
```
Each tier produces an array of `(input, output) => void` handlers; the matching OpenCode handler iterates and calls each in registration order.
Total: 52 base, 59 with team-mode. Each tier produces an object whose values are `(input, output) => void` handlers; the matching OpenCode handler invokes them in registration order via `safeHook()` wrappers.
## SUBSYSTEM INVENTORY