Adds a defensive tool-pair-validator hook that runs as the final step in
the messages transform pipeline. When compaction or context-window recovery
removes user messages containing tool_result blocks without removing the
preceding assistant message with tool_use blocks, this validator detects
the mismatch and either:
1. Injects missing tool_result parts into the next user message, or
2. Creates a synthetic user message with placeholder tool_results
This prevents Anthropic API errors like 'tool_use ids found without
tool_result blocks immediately after'.
Fixes#3014
Use the plugin session directory instead of process.cwd() when resolving project slash commands. This restores project and opencode-project slashcommand behavior when the runtime cwd differs from the actual session workspace.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Keep native /start-work resolvable on Sisyphus, but switch the work session back to Atlas when Atlas is registered. Stamp the outgoing agent with Atlas's actual list-display key so config→start-work execution resolves correctly and still falls back to Sisyphus when Atlas is unavailable.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
3 failures in fallback.cliproxyapi-matrix.test.ts:
- Root cause: leaked spyOn(getMainSessionID) in
tool-execute-before-session-notification.test.ts was never restored,
poisoning module state for subsequent tests in the same worker
- Added mockRestore() call and _resetModelFallbackForTesting in afterEach
3 failures in background-agent/manager.test.ts:
- Root cause: connected-providers-cache memConnected/memProviderModels
persisted across test files, making isReachable() skip fallback
candidates in retry tests
- Added mock.module for connected-providers-cache at file level
- Added _resetMemCacheForTesting export and global beforeEach cleanup
Full suite: 4638/4638 pass, 0 fail
Use `=== false` instead of falsy negation so that missing/undefined
fields are not treated as explicitly disabled. Only skip ~/.claude/rules
when all three flags are explicitly set to false.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When claude_code integration is disabled in oh-my-opencode.json, the
rules-injector now skips loading rules from ~/.claude/rules/ to prevent
Claude Code-specific instructions from leaking into non-Claude agents
(e.g., GPT-5.4 Sisyphus), which causes agent hallucination.
Also adds ~/.sisyphus/rules/ and ~/.opencode/rules/ as OpenCode-native
user rule directories that are always searched regardless of config.
Fixes#2920
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- background_output: snapshot read cursor before consuming, restore on
/undo message removal so re-reads return data (fixes#2915)
- MCP loader: preserve oauth field in transformMcpServer, add scope/
projectPath filtering so local-scoped MCPs only load in matching
directories (fixes#2917)
- runtime-fallback: add 'reached your usage limit' to retryable error
patterns so quota exhaustion triggers model fallback (fixes#2918)
Verified: bun test (4606 pass / 0 fail), tsc --noEmit clean
skill-context already filtered browser-related skills using the configured
browser provider, but the skill tool rebuilt discovery without forwarding
browserProvider. That caused skills like agent-browser to be prompt-visible
while skill() could still fail to resolve them unless browser_automation_engine.provider
was explicitly threaded through both paths.
Fix:
- pass skillContext.browserProvider from tool-registry into createSkillTool
- extend SkillLoadOptions with browserProvider
- forward browserProvider to getAllSkills()
- add regression tests for execution and description visibility
The skill tool previously only merged disk-discovered skills and preloaded
options.skills, so skills registered natively via ctx.skills (for example
from config.skills.paths or other plugins) were prompt-visible but not
discoverable by skill().
Fix:
- pass ctx.skills from tool-registry into createSkillTool
- extend SkillLoadOptions with nativeSkills accessor
- merge nativeSkills.all() results into getSkills()
- add test covering native PluginInput skill discovery
task_system now defaults to true instead of false. Users on opencode-go
or other plans were unable to invoke oracle or use subagent delegation
because experimental.task_system defaulted off.
The task system is the primary mechanism for oracle, explore, and other
subagent invocations — it should not be behind an experimental flag.
Users can still explicitly set experimental.task_system: false to opt out.
When a subagent session disappears from the status registry (process
crashed), the main agent was waiting the full stale timeout before
acting. Fix:
- Add sessionGoneTimeoutMs config option (default 60s, vs 30min normal)
- task-poller: use shorter timeout when session is gone from status
- manager: verify session existence when gone, fail crashed tasks
immediately with descriptive error
- Add legacy-plugin-toast hook for #2823 migration warnings
- Update schema with new config option
- Add legacy plugin startup warning when oh-my-opencode config detected
- Update CLI installer and TUI installer for new package name
- Split monolithic config-manager.test.ts into focused test modules
- Add plugin config detection tests for legacy name fallback
- Update processed-command-store to use plugin-identity constants
- Add claude-code-plugin-loader discovery test for both config names
- Update chat-params and ultrawork-db tests for plugin identity
Part of #2823
Sisyphus-authored fixes across 15 files:
- plugin-identity: align CONFIG_BASENAME with actual config file name
- add-plugin-to-opencode-config: handle legacy→canonical name migration
- plugin-detection tests: update expectations for new identity constants
- doctor/system: fix legacy name warning test assertions
- install tests: align with new plugin name
- chat-params tests: fix mock isolation
- model-capabilities tests: fix snapshot expectations
- image-converter: fix platform-dependent test assertions (Linux CI)
- example configs: expanded with more detailed comments
Full suite: 4484 pass, 0 fail, typecheck clean.
- Registry-based model family detection (provider-agnostic)
- Variant and reasoningEffort ladder downgrade logic
- Three-tier resolution: metadata override → family heuristic → unknown drop
- Comprehensive test suite covering all model families
Add support for object-style entries in fallback_models arrays, enabling
per-model configuration of variant, reasoningEffort, temperature, top_p,
maxTokens, and thinking settings.
- Zod schema for FallbackModelObject with full validation
- normalizeFallbackModels() and flattenToFallbackModelStrings() utilities
- Provider-agnostic model resolution pipeline with fallback chain
- Session prompt params state management
- Fallback chain construction with prefix-match lookup
- Integration across delegate-task, background-agent, and plugin layers
Reuse the stored model only for subsequent main-session messages when the UI provides no model, while preserving first-message behavior, explicit overrides, and subagent isolation.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
The existing normalizeToolArgSchemas only applies to omo plugin tools
(via tool-registry.ts), but MCP server tool schemas bypass this
sanitization entirely. MCP schemas with contentEncoding/contentMediaType
cause Gemini 400 errors.
Add sanitizeJsonSchema() to strip unsupported keywords from MCP tool
inputSchema before serialization in formatMcpCapabilities.
Fixes#2200
Supersedes #2666
YeonGyu-Kim