All versions since 6.38.0
6.38.0
Apr 16, 2026
Agent Communication Channels
The new catalyst-comms CLI gives Claude Code agents file-based communication across worktrees, sub-agents, and orchestrators without requiring servers or HTTP dependencies. Agents can join channels, send messages, poll for updates, and coordinate completion through simple bash commands that work with any agent workflow. Channel activity is logged locally at ~/catalyst/comms/ with automatic cleanup and human audit capabilities via catalyst-comms watch and status.
PRs
- dev: catalyst-comms — file-based agent communication channels (CTL-60) (#182) (51a73de)
- dev: worker polls until PR merges instead of exiting at pr-created (#180) (351cc95)
6.39.0
Apr 16, 2026
Orchestrator Launch Failure Detection
Workers that die immediately after dispatch (bad flags, environment errors) are now detected within 30 seconds instead of waiting 15 minutes for the stalled-worker detector. The orchestrator runs a batch health check after each dispatch wave, verifying worker PIDs and automatically flagging dead-on-arrival processes as failed with attention items.
PRs
6.39.1
Apr 16, 2026
Merged PR Status Writeback
The orchestration monitor now writes merged PR status back to worker signal files when it detects PRs have been merged on GitHub. Previously, merged PRs were only tracked in memory, causing the dashboard to show incorrect completion percentages when the orchestrator agent had already exited. Signal files now automatically update with status=done, phase=6, and merge timestamps for accurate project tracking.
PRs
7.0.0
Apr 16, 2026
Orchestration State Isolation
Orchestrator runtime state now lives in ~/catalyst/runs/<orch-id>/ instead of git worktrees, keeping your worktree clean during runs. Output files move to workers/output/ to reduce noise, while worker signal files stay in their expected locations. The monitor automatically handles both new runs-based and legacy worktree-based orchestrators.
PRs
- dev: decouple orch state from worktrees — runs/ dir (CTL-59) (#188)
- dev: decouple orch state from worktrees — runs/ dir (CTL-59) (#188) (a357eaa)
7.1.0
Apr 16, 2026
Session Resume Orchestration
When workers die mid-merge or stall with heartbeats, the orchestrator now revives them using claude --resume <session_id> instead of starting fresh — preserving full context while cutting costs ~10×. The system resolves session IDs from worker output streams and enforces per-ticket revive budgets, transitioning to stalled status when revival isn’t possible.
PRs
7.1.1
Apr 16, 2026
Worker Cost Display Fix
Dashboard now shows real-time worker costs (USD, input/output tokens, cache reads) instead of placeholder dashes. The orchestrator writes parsed usage data to each worker’s signal file, matching the existing global state format that powers the cost overview.
PRs
7.2.0
Apr 16, 2026
Authoritative Git and PR State
Workers are no longer marked as “stalled” based solely on signal file age — the orchestrator now uses git commit history and GitHub PR status as ground truth for completion detection. If a worker merged its PR but died before writing the terminal signal, it’s correctly recognized as complete rather than stalled. Run the orchestration monitor to see the improved accuracy in worker lifecycle tracking.
PRs
7.3.0
Apr 16, 2026
API Stream Idle Detection
Workers now recover immediately when hitting Claude API stream idle timeouts, instead of waiting up to 15 minutes for heartbeat staleness detection. Linear ticket states automatically transition when PRs are merged, with retroactive reconciliation available via orchestrate-bulk-close for tickets that stayed in “In Review” after successful merges.
PRs
- dev: detect API stream idle timeout in orchestrate-revive (CTL-62) (#196) (b89e342)
- dev: drive Linear ticket state transitions on PR merge (CTL-69) (#197) (dc58f32)
7.4.0
Apr 16, 2026
Auto-Dispatch Fixup Workers
Blocked PRs with unresolved review threads or failed checks now automatically trigger fixup workers after a 10-minute stabilization window, capping at 2 attempts before escalating to human attention. The orchestrator polls BLOCKED states alongside existing DIRTY/BEHIND handling, eliminating the need for manual intervention on stuck PRs. New signal tracking includes blockedSince, fixupAttempts, and lastFixupDispatchedAt for dashboard visibility.
PRs
7.5.0
Apr 18, 2026
Auto Orchestration & Warp Integration
Run catalyst orchestrate --auto N to automatically pick your top priority tickets from Linear, or use the new /catalyst-dev:setup-warp skill to generate terminal tab configs that launch orchestration sessions with proper naming and remote control. The --reuse-existing flag on worktree creation means your tab configs can safely reopen long-lived development environments without conflicts.
PRs
7.5.1
Apr 18, 2026
Warp Helper Script Bundling
The setup-warp skill now bundles its helper scripts directly in the plugin instead of referencing dotfiles that don’t exist for other users. Generated Warp tab configurations will point to the bundled open-project-tab.sh and trust-workspace.sh scripts, making the plugin work out of the box for everyone.
PRs
7.5.2
Apr 18, 2026
Warp Color Variant Fix
The setup-warp skill now only offers Warp’s 8 valid color variants (black, red, green, yellow, blue, magenta, cyan, white) instead of invalid options like purple and pink that caused Warp to reject generated tab configs on load.
PRs
7.5.3
Apr 19, 2026
Thoughts Symlink Protection
setup-catalyst no longer silently clobbers thoughts symlinks when repairing directory structure. The new catalyst-thoughts.sh helper detects when a regular directory has replaced an expected symlink and refuses to auto-fix, instead showing a recovery command to preserve any files written to the wrong location. Health checks now treat clobbered symlinks as fatal errors when humanlayer is configured.
PRs
7.5.4
Apr 19, 2026
Thoughts Profile Drift Repair
The catalyst-thoughts.sh init-or-repair command now automatically detects and fixes profile drift between your .catalyst/config.json and humanlayer’s repo mapping. When drift is detected, it runs humanlayer thoughts uninit --force followed by re-init with the correct profile and directory from your config. Previously, drift would cause silent failures that required manual intervention.
PRs
7.6.0
Apr 20, 2026
Worktree One-Shot Development Pipeline
Create a ticket worktree and launch Claude with the full research-to-ship pipeline pre-queued in one command. The new Warp tab variant runs /catalyst-dev:oneshot {{ticket}} automatically after worktree creation, enabling walk-away autonomous development workflows. Also adds PM kickoff prompts and fixes symlink preservation when copying plugin directories into new worktrees.
PRs
- dev: New Worktree One-Shot Warp variant (#215) (0614a96)
- dev: PM kickoff prompt + worktree symlink fix (#213) (b03fc87)
7.6.1
Apr 20, 2026
PM Parallel Orient Delegates
PM kickoff now dispatches three parallel sub-agents for orientation instead of running raw CLI fetches in the main context. This reduces a typical PM session start from ~15 tool calls with 30KB of JSON debris down to 3-4 clean tool calls, keeping the main context focused on PM reasoning rather than data collection.
PRs
7.7.0 Latest
Apr 21, 2026
Orchestrator Intelligence & Communications
Three productivity-focused areas land in the monitor: a restructured dashboard puts “what needs me?” first without scrolling, a new Comms view surfaces real-time catalyst-comms channels with live message feeds and cross-links from worker activity, and orchestrator briefings now include an auto-generated rollup that aggregates what shipped across all waves. Worker cost tooltips explain when metrics are unavailable, and PR status badges show merge conflicts or blocks at a glance across tables and cards.
PRs
- dev: briefing Sheet + orchestrator Briefing tab (CTL-105) (#234) (8324f1c)
- dev: catalyst-comms setup + website docs (CTL-113) (#231) (7ce31a4)
- dev: Comms view in orch-monitor (CTL-112) (#235) (ed8ba1a)
- dev: Comms view in orch-monitor (CTL-112) (#236) (81ef0e4)
- dev: dashboard IA three-zone layout (CTL-107) (#238) (9ce601f)
- dev: orchestrator rollup briefing (CTL-108) (#237) (007c8f3)
- dev: OTel health banner + worker cost tooltips (CTL-104) (#230) (0f63dbd)
- dev: persist orchestrator artifacts with hybrid archive (CTL-110) (#232) (003bce3)
- dev: PR status badges across orch-monitor (CTL-109) (#229) (b3510f8)
- dev: rename Process column to Worker, suppress dead PID on done workers (CTL-101) (#226) (6db8a76)
- dev: scaffold shadcn/ui interaction primitives in orch-monitor (CTL-97) (#223) (fbaba97)
- dev: setup-warp color-by-org convention, reserve blue for PM (#219) (4266d33)
- dev: wire catalyst-comms into orchestrate (CTL-111) (#222) (f1e0ecf)
- dev: worker + session drawers → shadcn Sheet (CTL-106) (#233) (6355968)
- dev: Active filter now hides done orchestrators (CTL-99) (#224) (39fbe22)
- dev: exclude abandoned workers from orch-monitor progress denominator (CTL-100) (#225) (0437300)
- dev: rename “Process died” → “Worker died” in attention feed (CTL-102) (#227) (5ee6c77)
- dev: TaskListSection empty/error states + worker-tasks debug endpoint (CTL-103) (#228) (8524299)