* feat(context-engine): add ContextEngine interface and registry
Introduce the pluggable ContextEngine abstraction that allows external
plugins to register custom context management strategies.
- ContextEngine interface with lifecycle methods: bootstrap, ingest,
ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn,
onSubagentEnded, dispose
- Module-level singleton registry with registerContextEngine() and
resolveContextEngine() (config-driven slot selection)
- LegacyContextEngine: pass-through implementation wrapping existing
compaction behavior for 100% backward compatibility
- ensureContextEnginesInitialized() guard for safe one-time registration
- 19 tests covering contract, registry, resolution, and legacy parity
* feat(plugins): add context-engine slot and registerContextEngine API
Wire the ContextEngine abstraction into the plugin system so external
plugins can register context engines via the standard plugin API.
- Add 'context-engine' to PluginKind union type
- Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy')
- Wire registerContextEngine() through OpenClawPluginApi
- Export ContextEngine types from plugin-sdk for external consumers
- Restore proper slot-based resolution in registry
* feat(context-engine): wire ContextEngine into agent run lifecycle
Integrate the ContextEngine abstraction into the core agent run path:
- Resolve context engine once per run (reused across retries)
- Bootstrap: hydrate canonical store from session file on first run
- Assemble: route context assembly through pluggable engine
- Auto-compaction guard: disable built-in auto-compaction when
the engine declares ownsCompaction (prevents double-compaction)
- AfterTurn: post-turn lifecycle hook for ingest + background
compaction decisions
- Overflow compaction: route through contextEngine.compact()
- Dispose: clean up engine resources in finally block
- Notify context engine on subagent lifecycle events
Legacy engine: all lifecycle methods are pass-through/no-op, preserving
100% backward compatibility for users without a context engine plugin.
* feat(plugins): add scoped subagent methods and gateway request scope
Expose runtime.subagent.{run, waitForRun, getSession, deleteSession}
so external plugins can spawn sub-agent sessions without raw gateway
dispatch access.
Uses AsyncLocalStorage request-scope bridge to dispatch internally via
handleGatewayRequest with a synthetic operator client. Methods are only
available during gateway request handling.
- Symbol.for-backed global singleton for cross-module-reload safety
- Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp)
- Set gateway request scope for all handlers, not just plugin handlers
- 3 staleness tests for fallback context hardening
* feat(context-engine): route /compact and sessions.get through context engine
Wire the /compact command and sessions.get handler through the pluggable
ContextEngine interface.
- Thread tokenBudget and force parameters to context engine compact
- Route /compact through contextEngine.compact() when registered
- Wire sessions.get as runtime alias for plugin subagent dispatch
- Add .pebbles/ to .gitignore
* style: format with oxfmt 0.33.0
Fix duplicate import (ControlUiRootState in server.impl.ts) and
import ordering across all changed files.
* fix: update extension test mocks for context-engine types
Add missing subagent property to bluebubbles PluginRuntime mock.
Add missing registerContextEngine to lobster OpenClawPluginApi mock.
* fix(subagents): keep deferred delete cleanup retryable
* style: format run attempt for CI
* fix(rebase): remove duplicate embedded-run imports
* test: add missing gateway context mock export
* fix: pass resolved auth profile into afterTurn compaction
Ensure the embedded runner forwards resolved auth profile context into
legacy context-engine compaction params on the normal afterTurn path,
matching overflow compaction behavior. This allows downstream LCM
summarization to use the intended provider auth/profile consistently.
Also fix strict TS typing in external-link token dedupe and align an
attempt unit test reasoningLevel value with the current ReasoningLevel
enum.
Regeneration-Prompt: |
We were debugging context-engine compaction where downstream summary
calls were missing the right auth/profile context in normal afterTurn
flow, while overflow compaction already propagated it. Preserve current
behavior and keep changes additive: thread the resolved authProfileId
through run -> attempt -> legacy compaction param builder without
broad refactors.
Add tests that prove the auth profile is included in afterTurn legacy
params and that overflow compaction still passes it through run
attempts. Keep existing APIs stable, and only adjust small type issues
needed for strict compilation.
* fix: remove duplicate imports from rebase
* feat: add context-engine system prompt additions
* fix(rebase): dedupe attempt import declarations
* test: fix fetch mock typing in ollama autodiscovery
* fix(test): add registerContextEngine to diffs extension mock APIs
* test(windows): use path.delimiter in ios-team-id fixture PATH
* test(cron): add model formatting and precedence edge case tests
Covers:
- Provider/model string splitting (whitespace, nested paths, empty segments)
- Provider normalization (casing, aliases like bedrock→amazon-bedrock)
- Anthropic model alias normalization (opus-4.5→claude-opus-4-5)
- Precedence: job payload > session override > config default
- Sequential runs with different providers (CI flake regression pattern)
- forceNew session preserving stored model overrides
- Whitespace/empty model string edge cases
- Config model as string vs object format
* test(cron): fix model formatting test config types
* test(phone-control): add registerContextEngine to mock API
* fix: re-export ChannelKind from config-reload-plan
* fix: add subagent mock to plugin-runtime-mock test util
* docs: add changelog fragment for context engine PR #22201
* feat(openai): add gpt-5.4 support and priority processing
* feat(openai-codex): add gpt-5.4 oauth support
* fix(openai): preserve provider overrides in gpt-5.4 fallback
* fix(openai-codex): keep xhigh for gpt-5.4 default
* fix(models): preserve configured overrides in list output
* fix(models): close gpt-5.4 integration gaps
* fix(openai): scope service tier to public api
* fix(openai): complete prep followups for gpt-5.4 support (#36590) (thanks @dorukardahan)
---------
Co-authored-by: Tyler Yust <TYTYYUST@YAHOO.COM>
createOllamaStreamFn() only accepted baseUrl, ignoring custom headers
configured in models.providers.<provider>.headers. This caused 403
errors when Ollama endpoints are behind reverse proxies that require
auth headers (e.g. X-OLLAMA-KEY via HAProxy).
Add optional defaultHeaders parameter to createOllamaStreamFn() and
merge them into every fetch request. Provider headers from config are
now passed through at the call site in the embedded runner.
Fixes#24285
* fix(gateway): flush throttled delta before emitChatFinal
The 150ms throttle in emitChatDelta can suppress the last text chunk
before emitChatFinal fires, causing streaming clients (e.g. ACP) to
receive truncated responses. The final event carries the complete text,
but clients that build responses incrementally from deltas miss the
tail end.
Flush one last unthrottled delta with the complete buffered text
immediately before sending the final event. This ensures all streaming
consumers have the full response without needing to reconcile deltas
against the final payload.
* fix(gateway): avoid duplicate delta flush when buffer unchanged
Track the text length at the time of the last broadcast. The flush in
emitChatFinal now only sends a delta if the buffer has grown since the
last broadcast, preventing duplicate sends when the final delta passed
the 150ms throttle and was already broadcast.
* fix(gateway): honor heartbeat suppression in final delta flush
* test(gateway): add final delta flush and dedupe coverage
* fix(gateway): skip final flush for silent lead fragments
* docs(changelog): note gateway final-delta flush fix credits
---------
Co-authored-by: Jonathan Taylor <visionik@pobox.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* feat(hooks): add trigger and channelId to plugin hook agent context
Adds `trigger` and `channelId` fields to `PluginHookAgentContext` so
plugins can determine what initiated the agent run and which channel
it originated from, without session-key parsing or Redis bridging.
trigger values: "user", "heartbeat", "cron", "memory"
channelId values: "telegram", "discord", "whatsapp", etc.
Both fields are threaded through run.ts and attempt.ts hookCtx so all
hook phases receive them (before_model_resolve, before_prompt_build,
before_agent_start, llm_input, llm_output, agent_end).
channelId falls back from messageChannel to messageProvider when the
former is not set. followup-runner passes originatingChannel so queued
followup runs also carry channel context.
* docs(changelog): note hook context parity fix for #28623
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(plugins): expose ephemeral sessionId in tool contexts for per-conversation isolation
The plugin tool context (`OpenClawPluginToolContext`) and tool hook
context (`PluginHookToolContext`) only provided `sessionKey`, which
is a durable channel identifier that survives /new and /reset.
Plugins like mem0 that need per-conversation isolation (e.g. mapping
Mem0 `run_id`) had no way to distinguish between conversations,
causing session-scoped memories to persist unbounded across resets.
Add `sessionId` (ephemeral UUID regenerated on /new and /reset) to:
- `OpenClawPluginToolContext` (factory context for plugin tools)
- `PluginHookToolContext` (before_tool_call / after_tool_call hooks)
- Internal `HookContext` for tool call wrappers
Thread the value from the run attempt through createOpenClawCodingTools
→ createOpenClawTools → resolvePluginTools and through the tool hook
wrapper.
Closes#31253
Made-with: Cursor
* fix(agents): propagate embedded sessionId through tool hook context
* test(hooks): cover sessionId in embedded tool hook contexts
* docs(changelog): add sessionId hook context follow-up note
* test(hooks): avoid toolCallId collision in after_tool_call e2e
---------
Co-authored-by: SidQin-cyber <sidqin0410@gmail.com>
* fix(hooks): deduplicate after_tool_call hook in embedded runs
(cherry picked from commit c129a1a74b)
* fix(hooks): propagate sessionKey in after_tool_call context
The after_tool_call hook in handleToolExecutionEnd was passing
`sessionKey: undefined` in the ToolContext, even though the value is
available on ctx.params. This broke plugins that need session context
in after_tool_call handlers (e.g., for per-session audit trails or
security logging).
- Add `sessionKey` to the `ToolHandlerParams` Pick type
- Pass `ctx.params.sessionKey` through to the hook context
- Add test assertion to prevent regression
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
(cherry picked from commit b7117384fc)
* fix(hooks): thread agentId through to after_tool_call hook context
Follow-up to #30511 — the after_tool_call hook context was passing
`agentId: undefined` because SubscribeEmbeddedPiSessionParams did not
carry the agent identity. This threads sessionAgentId (resolved in
attempt.ts) through the session params into the tool handler context,
giving plugins accurate agent-scoped context for both before_tool_call
and after_tool_call hooks.
Changes:
- Add `agentId?: string` to SubscribeEmbeddedPiSessionParams
- Add "agentId" to ToolHandlerParams Pick type
- Pass `agentId: sessionAgentId` at the subscribeEmbeddedPiSession()
call site in attempt.ts
- Wire ctx.params.agentId into the after_tool_call hook context
- Update tests to assert agentId propagation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
(cherry picked from commit aad01edd3e)
* changelog: credit after_tool_call hook contributors
* Update CHANGELOG.md
* agents: preserve adjusted params until tool end
* agents: emit after_tool_call with adjusted args
* tests: cover adjusted after_tool_call params
* tests: align adapter after_tool_call expectation
---------
Co-authored-by: jbeno <jim@jimbeno.net>
Co-authored-by: scoootscooob <zhentongfan@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
x-ai/grok models on OpenRouter do not support the reasoning.effort
parameter and reject payloads containing it with "Invalid arguments
passed to the model." Skip reasoning injection for these models, the
same way we already skip it for the dynamic "auto" routing model.
Closes#32039
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): support `thinkingDefault: "adaptive"` for Anthropic models
Anthropic's Opus 4.6 and Sonnet 4.6 support adaptive thinking where the
model dynamically decides when and how much to think. This is now
Anthropic's recommended mode and `budget_tokens` is deprecated on these
models.
Add "adaptive" as a valid thinking level:
- Config: `agents.defaults.thinkingDefault: "adaptive"`
- CLI: `/think adaptive` or `/think auto`
- Pi SDK mapping: "adaptive" → "medium" effort at the pi-agent-core
layer, which the Anthropic provider translates to
`thinking.type: "adaptive"` with `output_config.effort: "medium"`
- Provider fallbacks: OpenRouter and Google map "adaptive" to their
respective "medium" equivalents
Closes#30880
Made-with: Cursor
* style(changelog): format changelog with oxfmt
* test(types): fix strict typing in runtime/plugin-context tests
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
