- Clear stop state when user sends new message (chat.message handler)
- Add isContinuationStopped check to session error recovery block
- Continuation resumes automatically after user interaction
- BackgroundManager.shutdown() now aborts all running child sessions via
client.session.abort() before clearing state, preventing orphaned
opencode processes when parent exits
- Add onShutdown callback to BackgroundManager constructor, used to
trigger TmuxSessionManager.cleanup() on process exit signals
- Interactive bash session hook now aborts tracked subagent opencode
sessions when killing tmux sessions (defense-in-depth)
- Add 4 tests verifying shutdown abort behavior and callback invocation
Closes#1240
* feat(delegate-task): add prometheus self-delegation block and delegate_task permission
- Block prometheus from delegating to itself via delegate_task
- Grant delegate_task permission to prometheus when called as subagent
- Other subagents still have delegate_task disabled
* feat(version): add OPENCODE_NATIVE_AGENTS_INJECTION_VERSION constant
* docs: add deprecation notes for directory-agents-injector
* feat(hooks): auto-disable directory-agents-injector for OpenCode 1.1.37+
---------
Co-authored-by: justsisyphus <justsisyphus@users.noreply.github.com>
/review command and some Claude Code built-in commands trigger
tool.execute.after hooks with undefined output, causing crashes
when accessing output.metadata or output.output.
Fixes#1035
Co-authored-by: sisyphus-dev-ai <sisyphus-dev-ai@users.noreply.github.com>
Show warning toast when hasConnectedProvidersCache() returns false,
indicating model filtering is disabled. Prompts user to restart
OpenCode for full functionality.
- Add permission: [{ permission: 'question', action: 'deny' }] to session.create()
in background-agent and delegate-task for SDK-level blocking
- Add subagent-question-blocker hook as backup layer to intercept question tool
calls in tool.execute.before event
- Ensures subagents cannot ask questions to users and must work autonomously
When AI generates AskUserQuestion tool calls with option labels longer
than 30 characters, opencode validation rejects them with "too_big" error.
This fix adds a pre-tool-use hook that automatically truncates labels
to 30 characters (with "..." suffix) before the validation occurs.
Fixes the error:
"The question tool was called with invalid arguments: expected string
to have <=30 characters"
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The background_task config (providerConcurrency, modelConcurrency, etc.)
was not being passed to BackgroundManager, causing all models to use
the hardcoded default limit of 5 instead of user-configured values.
- Changed fetchAvailableModels to read from ~/.cache/opencode/models.json
- Prevents plugin startup hanging caused by SDK client.config.providers() call
- Updated doctor model-resolution check to show available models from cache
- Added cache info display: provider count, model count, refresh command
- Remove Step 3 in model-resolver that forced first fallbackChain entry
even when unavailable, blocking system default fallback
- Add sisyphusJuniorModel option to delegate_task so agents["Sisyphus-Junior"]
model override is respected in category-based delegation
- Update tests to reflect new fallback behavior
Apply case-insensitive utilities across agents, delegation, and tool
restrictions. Removes duplicate implementations from agents/utils.ts.
Agent validation now normalizes input to canonical names.
Bug: When switching from Prometheus to Sisyphus, the Prometheus ultrawork
template was still injected because:
1. setSessionAgent() only sets on first call, ignoring subsequent updates
2. keyword-detector relied solely on input.agent which could be stale
Fix:
- Use updateSessionAgent() instead of setSessionAgent() in index.ts
- keyword-detector now uses getSessionAgent() as primary source, fallback to input.agent
- Added tests for agent switch scenario
The prometheus-md-only hook was not enforcing file restrictions because
getSessionAgent() returned undefined - setSessionAgent was only called
in message.updated event which doesn't always provide agent info.
- Add setSessionAgent call in chat.message hook when input.agent exists
- Add session state tests for setSessionAgent/getSessionAgent
- Add clearSessionAgent cleanup to prometheus-md-only tests
This ensures prometheus-md-only hook can reliably identify Prometheus
sessions and enforce .sisyphus/*.md write restrictions.
- Delete src/hooks/preemptive-compaction/ entirely
- Remove dcp_for_compaction from schema and executor
- Clean up related imports, options, and test code
- Update READMEs to remove experimental options docs
The watermark (commit footer and co-author) was inconsistently applied because:
1. The skill tool didn't receive gitMasterConfig
2. The approach was 'default ON, inject DISABLED override' which LLMs sometimes ignored
This refactors to 'inject only when enabled' approach:
- Remove hardcoded watermark section from base templates
- Dynamically inject section 5.5 based on config values
- Default is still ON (both true when no config)
- When both disabled, no injection occurs (clean prompt)
Also fixes missing config propagation to skill tool and createBuiltinAgents.
Background tasks were defaulting to $HOME instead of the parent session's
working directory. This caused background agents to scan the entire home
directory instead of the project directory, leading to:
- High CPU/memory load from scanning unrelated files
- Permission errors on system directories
- Task failures and timeouts
The fix retrieves the parent session's directory before creating a new
background session and passes it via the query.directory parameter.
Files modified:
- manager.ts: Look up parent session directory in launch()
- call-omo-agent/tools.ts: Same fix for sync mode
- look-at/tools.ts: Same fix for look_at tool
- sisyphus-task/tools.ts: Same fix + interface update for directory prop
- index.ts: Pass directory to sisyphusTask factory
Allow optional model variant config for agents and categories.
Propagate category variants into task model payloads so
category-driven runs inherit provider-specific variants.
Closes: #647
- Revert messages.transform-only approach (experimental hook unreliable)
- Inject context in chat.message after claudeCodeHooks runs
- Order: keywordDetector → claudeCodeHooks → contextInjector
- Works independently of claude-code-hooks being enabled/disabled
- Ultrawork content now reliably injected to model
- Remove contextInjector call from chat.message hook chain
- Context injection now only happens in messages.transform hook
- This ensures UI displays original user message while model receives prepended context
- Fixes bug where commit message promised clone behavior but implementation mutated directly
- keyword-detector now registers keywords to ContextCollector
- context-injector consumes and injects via chat.message hook
- Removed keyword detection logic from claude-code-hooks
- Hook order: keyword-detector → context-injector → claude-code-hooks
- ultrawork now works even when claude-code-hooks is disabled
* fix(lsp): cleanup orphaned LSP servers on session.deleted
When parallel background agent tasks complete, their LSP servers (for
repos cloned to /tmp/) remain running until a 5-minute idle timeout.
This causes memory accumulation with heavy parallel Sisyphus usage,
potentially leading to OOM crashes.
This change adds cleanupTempDirectoryClients() to LSPServerManager
(matching the pattern used by SkillMcpManager.disconnectSession())
and calls it on session.deleted events.
The cleanup targets idle LSP clients (refCount=0) for temporary
directories (/tmp/, /var/folders/) where agent tasks clone repos.
* chore: retrigger CI checks
justsisyphus