Sidecar lifecycle:
- Move startSideCar() out of register() into an api.on("gateway_start", ...)
handler. register() runs in every CLI subprocess that loads plugins
(e.g. `openclaw completion`, `openclaw doctor`); eagerly spawning a
long-lived process there hung `openclaw update`'s post-update steps.
- Spawn the sidecar with detached: true, stdio routed to a log file fd,
and call .unref() so the host's event loop is never held by the child.
Even if a future caller invokes startSideCar in a non-gateway context,
it can no longer block that host from exiting.
- Sidecar logs now go to ~/.openclaw/logs/dirigent-sidecar.log instead of
being piped through the host logger.
Plugin SDK convention update:
- Wrap default export with definePluginEntry({ id, name, description, register })
per the current openclaw plugin authoring contract.
- Switch all imports from the deprecated root barrel "openclaw/plugin-sdk"
to focused subpaths "openclaw/plugin-sdk/core" and
"openclaw/plugin-sdk/plugin-entry".
- Modernize openclaw.plugin.json: drop entry/version, add description,
declare contracts.tools[] for the 6 tools, set activation.onStartup: true
so gateway_start fires for this plugin at boot.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two bugs that prevented turn-manager dormancy from ever triggering:
1. isEmptyTurn too strict: agents output multi-line text ending with
"NO_REPLY" on the last line, but the regex ^NO_REPLY$ required the
entire string to match. Now checks only the last non-empty line.
2. blocked_pending counter inflation: non-speaker suppressions incremented
the counter but their stale NO_REPLYs were discarded at the
!isCurrentSpeaker early return without decrementing. Over a full cycle
the counter inflated by the number of suppressions, causing the agent's
real empty turn to be misidentified as stale when it finally arrived.
Fix: at both early-return points in agent_end (!isCurrentSpeaker and
!isTurnPending), drain blocked_pending when the turn is empty.
Also fixed: pollForTailMatch now uses any-message detection (instead of
tail-fingerprint content matching) with a 30 s timeout, avoiding infinite
polling when agents send concise Discord messages after verbose LLM output.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two bugs in concluded discussion channel handling:
1. before_model_resolve did not check rec.discussion.concluded, so it
still initialized the speaker list and ran turn management. Fixed by
returning NO_REPLY early for concluded discussions (same as report mode).
2. message_received fired for all agent VM contexts, causing multiple
"This discussion is closed" auto-replies per incoming message. Fixed
with process-level dedup keyed on channelId:messageId (same pattern
as agent_end runId dedup). Also fixed message_id extraction to look
in metadata.conversation_info.message_id first.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Complete rewrite of the Dirigent plugin turn management system to work
correctly with OpenClaw's VM-context-per-session architecture:
- All turn state stored on globalThis (persists across VM context hot-reloads)
- Hooks registered unconditionally on every api instance; event-level dedup
(runId Set for agent_end, WeakSet for before_model_resolve) prevents
double-processing
- Gateway lifecycle events (gateway_start/stop) guarded once via globalThis flag
- Shared initializingChannels lock prevents concurrent channel init across VM
contexts in message_received and before_model_resolve
- New ChannelStore and IdentityRegistry replace old policy/session-state modules
- Added agent_end hook with tail-match polling for Discord delivery confirmation
- Added web control page, padded-cell auto-scan, discussion tool support
- Removed obsolete v1 modules: channel-resolver, channel-modes, discussion-service,
session-state, turn-bootstrap, policy/store, rules, decision-input
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When the turn manager determines it's an agent's turn (checkTurn.allowed),
the rules engine's evaluateDecision() could still override the model to
no-reply with reason 'rule_match_no_end_symbol'. This happened because:
1. The sender of the triggering message (another agent) was not in the
humanList, so the rules fell through to the end-symbol check.
2. getLastChar() operates on the full prompt (including system content
like Runtime info), so it never found the end symbol even when the
actual message ended with one.
Fix: return early from before_model_resolve after the turn check passes,
skipping the rules-based no-reply override entirely. The turn manager is
the authoritative source for multi-agent turn coordination.
Tested: 3-agent counting chain ran successfully (3→11) with correct
NO_REPLY handling when count exceeded threshold.
