v0.10.2

Full Changelog: v0.10.1...v0.10.2

v0.10.1

Full Changelog: v0.10.0...v0.10.1

v0.10.0

Full Changelog: v0.9.1...v0.10.0

v0.9.1

Full Changelog: v0.9.0...v0.9.1

v0.9.0

Skill Bill v0.9.0 ships a redesigned desktop app where you can start work by chatting with an agent. Goals now run in their own temporary worktrees, and Skill Bill state moves out of your repositories.

Highlights

A new desktop app

The redesigned app replaces the old tray app. It has a work board, run details, a sidebar, and a configuration pane.

• New Task home screen. Describe the work in a chat, pick the repository and agent, and start a goal from the conversation. Each goal gets a descriptive title. Follow-up messages queue while a reply is still streaming.
• Agent chat. Chat with an execution's agent from a side pane. You can open several chats per execution, paste images, and use Stop and Send now in every composer. Every chat knows Skill Bill and can start runs.
• Embedded terminal. Terminals with multiple tabs, each starting in the selected project.
• Spec browser. Browse spec bundles, open their files from goal details, and delete bundles from the Specs tree.
• Changes window. A read-only diff of a goal's changed files, for the whole goal or a single subtask.
• Subtask inspector. Review passes, audit results, and a live QA checklist for the selected subtask.
• Run favorites. Pin runs and reach them from their own sidebar section.

Goals stay out of your checkout

• Temporary worktrees. Each new goal runs in its own worktree, which is removed when the goal finishes. Your checkout stays untouched.
• Project store. State, specs, and per-phase instruction files now live in a per-project store under ~/.skill-bill instead of inside your repository.

Broader QA that never blocks

• phase:qa now tests backend endpoints and web journeys as well as desktop apps, devices, and simulators.
• QA never blocks a run. Every QA failure is recorded, reported to the desktop app, and the run continues.

Workflow improvements

• Goal assessment. After all subtasks finish, the whole goal is assessed and the issues found are fixed. Audits check criteria against each other and check whether the tests actually prove anything.
• Verified standalone review. Findings are verified before they are shown. Each finding carries its verifier status, and the findings are kept when verification blocks.
• Branch monitoring. phase:monitor branch:<name> watches and repairs a branch that has no open pull request.
• Reusable keys. Start a fresh goal on a key whose earlier goal is completed or cancelled.
• goal complete. Mark a goal complete when its work landed outside the runtime.
• Fresh goal quality. Choose to rerun a goal's quality checks from scratch.
• Better pull requests. Goal PRs follow the repository's PR template and get clear titles and descriptions.
• Claude account selection. Desktop runs use the Claude account chosen in settings.

Fixes

• The desktop app gets your full shell environment when launched from the Dock, not just a default PATH.
• Boards and run details no longer stall during active goals, and status reads are faster.
• The Claude chat process stays alive between turns, and the running state updates when a run stops.
• When the stacked base branch is gone, goal PRs open against the default branch. A goal's own branch is force-pushed when integrating origin fails.
• Goal finish stays pending while subtasks are still running. The finish monitor passes once checks are green on the current PR head.
• Cursor review and goal-assessment output no longer drop verdict and finding lines.
• Every verified finding is repaired whatever its severity, and only material gaps count against a run. Audits accept shorthand citations and extra prose.
• Validation can rebuild artifacts, repair build failures, and run in nested Gradle projects. Checks whose host tooling is missing are skipped with a warning, and gate commands that fail to start are reported.
• Retrying a blocked phase keeps goal artifacts. Interrupted plans resume from their checkpoints, and long plans no longer hit a planning time budget.
• Codex sessions continue after running out of output tokens.
• Busy SQLite write transactions are retried instead of failing.
• QA tests the intake's requirements on the checked-out tree and skips only when QA tools are unavailable.
• Repository identities with Windows drive paths are accepted.
• Running the installer from the desktop app's terminal restarts the desktop app.

Upgrade notes

• The old desktop tray app is removed. The new desktop app is installed in its place.
• New goals run in temporary worktrees. Work in progress in your own checkout is no longer touched by a running goal.
• Skill Bill state now lives in the per-project store under ~/.skill-bill. Run skill-bill project path from a repository root to find it.

Full changelog: v0.8.0...v0.9.0

v0.8.0

Skill Bill v0.8.0 (https://github.com/Sermilion/SkillBill/releases/tag/v0.8.0) adds mobile QA, guided recovery,
and criterion-by-criterion implementation.

• Run Android and iOS checks on devices or simulators with phase:qa.
• Run QA on the combined result after all subtasks finish, with isolated worktrees and repair attempts.
• Inspect saved diagnoses and supported recovery actions for blocked runs in the terminal and desktop app.
• Generate sanitized defect reports from evidence collected during blocked runs.
• Add project-specific phase instructions through .skill-bill/phases/.md.
• Implement and independently verify each acceptance criterion before advancing. Repair attempts retain separate
budgets for each criterion.

• Monitor and repair multiple pull requests in separate temporary worktrees.
• Use one planning flow for single-subtask features and larger goals.
• View review findings, audit results, QA reports, and finalization progress in the desktop app.

The desktop app replaces the IntelliJ plugin and VS Code extension. Installation removes previously installed
copies.

This release also includes fixes for:

• Resuming blocked, failed, interrupted, and cancelled goals without losing planning progress.
• Saving completed subtask plans before peer planners finish.
• Showing the current phase and agent, and detecting crashed runners.
• Preserving verification checkpoints when recording failures.
• Retaining repair budgets across restarts and counting unsuccessful attempts correctly.
• Running the full validation gate for every goal subtask.
• Handling moving PR branches, merge conflicts, and CI repair limits.
• Preserving databases, backups, QA workspaces, and app state during upgrades.
• Keeping archived specs out of active planning.
• macOS and Windows release packaging, plus build checks before standard releases.

v0.7.0

Full Changelog: v0.6.1...v0.7.0

v0.6.1

Full Changelog: v0.6.0...v0.6.1

v0.6.0

Full Changelog: v0.5.3...v0.6.0

v0.5.3

Full Changelog: v0.5.1...v0.5.3