v0.14.2

Release Notes (v0.13.0 → v0.14.2)

🎉 New Features

Independent Per-Phase Verification with Must-Haves (#1024)

• Planners now attach explicit "must-haves" (invariants) to each implementation phase, and a dedicated phase-verifier agent independently checks that they're actually met in the worktree after the phase completes
• The verifier gathers its own evidence — reading files and running targeted tests — and returns a GO/NO-GO verdict per must-have, reporting violations without silently fixing them
• The orchestrator now runs per-phase verification and code review after each phase instead of only at the end, catching drift from the plan earlier

Aggressive Parallelization for Complex Issues (#1018)

• Complex issues now execute implementation steps in parallel waves by default, dramatically reducing completion time
• The planner uses contract-based parallelism — when two steps share a type or interface, it defines the contract in both and runs them concurrently instead of sequentially
• Compile errors between parallel steps are expected and handled gracefully; a post-wave validation checkpoint verifies the integrated result before moving on

Background Workflow Agents (#1020)

• All phase agents (enhancer, analyzer, planner, implementer, artifact reviewer) now run in the background by default, freeing the orchestrator context window for ongoing conversation
• Code reviewer remains in foreground as it requires direct tool access

Smart Rebase/Merge Strategy (#1017)

• il rebase and il finish now automatically detect when rebasing would be problematic and fall back to merging from the parent branch instead
• Triggers when the loom branch already has merge commits from the parent, or when more than 20 commits would be rebased (configurable via rebase.maxCommitsForRebase)

Smarter PR Review Resolution (#1021)

• Agents now cross-reference review submissions (the "Submit review" verdicts) against inline comments to identify truly unresolved feedback, instead of re-reporting findings that were already addressed
• Agents now reply directly within specific comment threads rather than posting standalone reviews

Smarter Planning with Dedicated Research Agent (#879, #880)

• Plan mode delegates research to the Analyze agent instead of doing inline discovery, producing deeper and more thorough implementation plans
• Plans benefit from the same structured research methodology used during issue implementation

Integration-Point Must-Haves in Plans (#912)

• The planner detects producer/consumer relationships between parallel child issues in the same wave and generates substantive must-haves on verification issues
• Ensures the verifier checks that consumers actually use real instances instead of null placeholders after parallel work merges

Composable Automation Flags for il plan (#908, #909)

• --yolo is decomposed into independent, composable flags so you can fine-tune how much autonomy the planner gets
• --one-shot=noReview skips confirmation gates but keeps permission prompts; --one-shot=bypassPermissions skips both
• --dangerously-skip-permissions is available standalone on both il plan and il start; --auto-swarm is independent; --yolo remains a convenient superset shorthand

Configurable Effort Levels (#959)

• New --effort flag (low / medium / high / max) on il start, il spin, and il plan controls how deeply agents reason through each phase
• Resolution chain: CLI flag > loom metadata > settings > environment variable > default; swarm mode uses per-agent defaults and supports separate swarmEffort overrides

Cross-Repo Epic Planning (#1008)

• il plan now guides you through decomposing work that spans multiple repositories, coordinating via shared contracts rather than cross-repo blocking dependencies

Epic Swarm Completion Reports (#928, #934)

• When a swarm finishes, iloom generates a structured report covering what each child issue delivered, integration points to review carefully, and lessons learned — posted as a comment with collapsible per-issue sections
• Includes testing guidance and a verification checklist derived from actual implementation data, not boilerplate

Swarm Visibility Improvements (#901, #903, #950)

• Full agent routing table displayed after a swarm completes so you can see how every issue was handled
• Comment strategy switched to a single final summary; epic dev server port forwarded to the orchestrator for web output verification after merging child work

Wave Verifier Enhancements (#899, #902, #911)

• Verification results posted as comments directly on the verifier's issue, creating a visible audit trail
• Code review step added to wave verification, with per-file parallel fixes for critical findings

Authenticated Image Support in Linear Issues (#1005)

• Images embedded in Linear issues (uploads.linear.app) and GitHub user-attachments are now automatically downloaded, cached, and made viewable by agents, so they can analyze screenshots, mockups, and diagrams attached to issues

1M Context Window Model Support (#907)

• opus[1m] and sonnet[1m] model selections are available across all commands
• Swarm orchestrator defaults to opus[1m] in Balanced and Maximum Quality modes for deeper reasoning over large codebases

Dev Server TUI with Sticky Status Bar (#922, #923, #955)

• il dev-server shows a persistent status bar with the external URL, container status, and port mapping
• Keyboard shortcuts: o (open browser), c (copy URL), r (restart), q (quit) — activates automatically in TTY environments with zero new dependencies

Docker BuildKit Secrets and Runtime Environment Variables (#919, #935)

• New dockerBuildSecrets config maps secret IDs to local files, passed as --secret flags during docker build
• New dockerRunEnv setting and repeatable -e, --env KEY=VALUE CLI flag on dev-server, open, and run for passing environment variables to containers at runtime (precedence: settings < .env files < CLI flags)

Configurable Dev Server Protocol (#921)

• New protocol option (http/https) in web capabilities settings so dev server URLs display the correct scheme across il open, il run, and il dev-server

Docker Dev Server Detection (#964)

• il init detects Dockerfiles in your project and offers to configure Docker-based dev servers automatically

Configurable Default Labels (#990)

• New issueManagement.github.defaultLabels and issueManagement.jira.defaultLabels settings automatically apply labels to issues created by iloom agents

Structured Draft PR Bodies (#953, #954)

• Draft PRs created by il start include the issue title as a heading and the issue body in a collapsible details section, giving reviewers immediate context

Streamlined VS Code Init Wizard (#952)

• When running from VS Code, the config wizard detects the IDE automatically and skips the IDE selection question

Code Review Fix Tracking (#961)

• New fix recap entry type lets agents log when a review finding has been addressed, creating a clear audit trail from finding to resolution

Pre-Trusted Tools in Plan Mode (#869, #900)

• Safe, read-only tools and issue management tools are pre-trusted during planning, reducing approval prompts so you can focus on reviewing the plan

Auto-Ignore .env.local Files (#927)

• Local environment override files (.env.local, .env.development.local, etc.) are automatically added to the global gitignore on upgrade, preventing accidental commits of secrets

🐛 Bug Fixes

Session Summary Works Under Enterprise Accounts (#1026)

• Session summary generation no longer hard-fails on Enterprise Account setups where bare mode is incompatible with the account's authentication — it now falls back gracefully like other steps, instead of erroring out with "Invalid API key" (fixes #1022)

Docker Cleanup No Longer Hangs (#1024)

• Docker CLI calls during cleanup (docker ps, docker rm) now have timeouts, so an unresponsive Docker daemon no longer causes cleanup to hang indefinitely

--create-only No Longer Fails When IDE Is Missing (#1023)

• il start --create-only no longer requires the configured IDE (VS Code, Cursor, etc.) to be installed, since it never launches it — matching the behavior of --no-code

Draft PR Finish Flow (#1019)

• il finish now handles closed issues gracefully when the loom has a draft PR, and skips straight to cleanup when the draft PR is already merged

Settings Validation (#1014, #1015)

• Global settings with partial configs (e.g., shared API token without team ID) are no longer incorrectly rejected, and full Claude model IDs like claude-opus-4-6[1m] are now accepted

PR Finish on Linear/Jira Projects (#983)

• il finish <pr-number> now works correctly on projects using Linear or Jira for issue tracking

Plan Agent Respects Configured Issue Tracker (#904)

• The planner knows which issue tracker and VCS provider are configured, preventing it from using gh CLI commands when the tracker is Linear or Jira

Clarifying Questions Batched (#914)

• The planner batches up to 4 independent clarifying questions per round instead of asking one at a time

Cross-Platform & Install Fixes (#982, #1006, #1003)

• Windows compatibility fix for temporary directory paths (#982)
• il update detects Volta-managed installs and uses volta install instead of npm install -g (#1006)
• Fixed "Unknown system error" crash when the architect prompt exceeded macOS argument size limits (#1003)

Dev Server Fixes (#924, #926, #955, #958)

• Main worktree now correctly uses the base port (default 3000) instead of a hash-derived offset (#924)
• TUI dev server respects custom dev commands from config, and Docker container name conflicts are resolved for concurrent looms (#926)
• The [r] restart key in the TUI dev server now actually restarts the server (#955)
• il list --json correctly detects capabilities for the main worktree (#958)

Swarm & Rebase Fixes (#925, #949)

• Read-only tools are pre-granted during rebase conflict resolution, so agents don't prompt for every file read (#925)
• Wave verification tasks and swarm workers correctly route recap entries to child recap files instead of the epic's (#949)

Other Fixes (#884, #897, #898, #910, #913, #956, #1002)

• Creating a loom from VS Code with a short title and description body no longer fails with "Invalid branch name" (#884)
• Cleanup finds swarm child worktrees using both issue-{id} and issue/{id} branch patterns (#897) and handles Linear/Jira-style IDs like RANDOM-13 (#898)
• il list from a child worktree overriding mainBranch no longer filters out all looms (#910)
• il start in branch mode reuses an existing local branch instead of failing (#913)
• Documentation files in the agents directory no longer prevent agents from loading (#956)
• VS Code extension shows the correct Linear icon for Linear-tracked issues (#1002)

---

Full Changelog: v0.13.0...v0.14.2