v0.11.0

Release Notes (v0.10.0 → v0.11.0)

🎉 New Features

Linux, WSL, and tmux Terminal Support

• iloom now works on Linux with GUI terminals (gnome-terminal, konsole, xterm), WSL via Windows Terminal, and headless environments via tmux fallback (#796)
• Auto-detection: tries GUI terminals first, falls back to tmux when no display is available
• Unblocks il start on headless Linux environments (SSH, Docker, Code Server) where iloom was previously unusable
• Thanks to @TickTockBent for the implementation and @rexsilex for the original architecture

Auto-Swarm: One Command from Plan to Execution

• il plan --auto-swarm chains the entire epic lifecycle — plan decomposition, epic workspace creation, and swarm launch — into a single unattended pipeline (#769)
• If the plan produces no child issues, it falls back to a normal autonomous loom
• VS Code extension can own the harness socket to manage the pipeline externally

PR Code Reviews with Inline Comments

• Agents can now post inline code review comments directly on GitHub pull requests, addressing specific lines of code rather than leaving generic top-level feedback (#819)
• Severity-based verdicts: Critical and Warning findings trigger REQUEST_CHANGES, suggestions are informational
• In one-shot mode, agents automatically triage and address existing review comments — fixing quick and moderate items, deferring complex or risky ones
• When opening a PR, agents check for unresolved review comments and work through them

Swarm Quality Presets in Configuration Wizard

• il init now offers swarm quality modes upfront: Maximum Quality (Opus everywhere), Balanced (Opus for analysis and workers, Sonnet for planning), and Fast & Cheap (Sonnet/Haiku) (#731, #777)
• Pick a preset and start swarming without manually configuring per-agent models

--complexity Flag to Skip Complexity Evaluation

• Tell iloom exactly how complex a task is instead of waiting for the evaluator: il start --complexity=simple or il spin --complexity=complex (#816)
• On il start, the override persists in loom metadata for future spins
• On il spin, it applies to the current session only

Configurable Per-Agent Swarm Models

• Set different AI models for each phase agent in swarm mode via per-agent swarmModel overrides, independent of non-swarm model settings (#736)

Post-Swarm Code Review Phase

• After all swarm child agents complete and merge, the orchestrator automatically runs a code review on the merged epic branch and auto-fixes high-confidence findings (#827)
• Configurable via spin.postSwarmReview setting (enabled by default)
• Non-blocking: review or fix failures don't prevent finalization
• Single-pass constraint prevents infinite review-fix loops

--create-only Flag for Workspace-Only Creation

• New convenience flag on il start that bundles --no-claude --no-code --no-terminal --no-dev-server — creates the isolated workspace without launching anything (#825)
• Useful for automation, scripting, and cases where you want to set up the environment before starting work

--json-stream for Headless Automation

• il rebase, il finish, and il commit now support --json-stream for JSONL streaming to stdout with progress on stderr — useful for CI pipelines and tooling integrations (#635)

il install-deps Command

• New utility command that detects and installs dependencies using iloom config files or lockfile detection, with support for pnpm, npm, and yarn (#780)

🐛 Bug Fixes

Swarm Reliability and Performance — Workers now produce descriptive commit messages instead of generic "fixes #NNN"; dependencies are installed after merging worker branches so subsequent workers have up-to-date packages; re-spinning skips done child issues so partial failures don't redo completed work; reduced per-agent overhead keeps worker context windows lean; il spin in child worktrees now shows a clear error instead of conflicting with the orchestrator (#781, #831, #830, #789, #792, #774)

Agent Analysis Quality — Anti-confabulation rules prevent agents from fabricating cross-component values; anti-parroting rules ensure assumptions add genuine insight rather than restating the issue description; recap entries filter out agent infrastructure noise (#823, #828, #817)

Jira Wiki Markup Sanitization — Prevents Jira Wiki format from leaking into issue comments when Jira is configured as the issue tracker (#728)

Code Reviewer Works on All Branch Types — Previously only worked with uncommitted changes. Now diffs against the appropriate merge base for feature branches (#785)

Error Visibility When Claude Launch Fails — Error output from Claude is now properly surfaced instead of being silently swallowed (#786)

---

Full Changelog: v0.10.0...v0.11.0