{
  "latest": "0.7.1",
  "channel": "early-access",
  "updatePolicy": "OpenThunder updates itself. New versions download in the background and install the next time you quit, or right away if you choose Restart now. The upgrade works from any older version to the newest.",
  "releases": [
    {
      "version": "0.7.1",
      "date": "2026-10-21",
      "title": "Clearer, and Rules your whole team and its agents follow",
      "notes": [
        "Clearer everywhere. The sidebar now has six places: Check, Changes, Rules, Proof, Tasks, Settings, with everything else under More tools. First run is one screen: choose a repo folder and you land on its Check. Every verdict reads Ready to merge, Needs review, or Blocked, with the reason, and failures now say what happened and what to do next.",
        "Team Rules you can trust. The gate reads Rules from the branch you merge into, so a change (or an AI agent) can't loosen the Rules that judge it. Rules can require that tests and the security scan actually ran and passed for changes that reach widely used code; a check that didn't run counts as not proven.",
        "Rules for your whole org and your agents. Teams can set Rules once in OpenThunder Cloud and enforce them in every repo's CI (openthunder cloud rules pull). openthunder rules compile puts them into Claude Code's own settings, so locked paths are blocked before an edit happens, and into AGENTS.md for other agents.",
        "Try a Rule before you enforce it. The new Rules page (and openthunder rules preview) shows what a rule would have done to your recent changes, for example 'would have blocked 4 of the last 212', before anyone turns it on.",
        "Fixed: Night Shift's run queue now works on OpenThunder Cloud (Postgres); it previously failed to start runs there."
      ]
    },
    {
      "version": "0.7.0",
      "date": "2026-10-10",
      "title": "Night Shift: put AI agents to work, wake up to verified results",
      "notes": [
        "New: Night Shift. Queue engineering tasks and research questions, and OpenThunder's agents work them while you are away, each in its own isolated workspace. You wake up to a ranked list, what is ready to ship, what needs a decision, what is blocked, with the evidence behind every verdict. Nothing is merged without you.",
        "Two kinds of agent. Development agents write and verify code changes through an independent review loop. Research agents answer a question with a cited brief, and every citation is fetched and independently checked, so a source that does not actually back the claim is flagged, not trusted.",
        "Runs survive restarts. A crash or a quit no longer loses in-flight work: Night Shift picks up where it left off, so overnight really means overnight.",
        "As always, the agent proposes and OpenThunder verifies: the ship, caution, or hold verdict is computed from recorded executions, never an agent's say-so."
      ]
    },
    {
      "version": "0.6.7",
      "date": "2026-09-20",
      "title": "Goals, and a clearer way around the app",
      "notes": [
        "New: Goals. Set an outcome you care about (ship a feature, complete a course, pass a certification) and OpenThunder tracks progress from executed evidence rather than guesswork: a mission that ran no checks is marked needs-evidence, never counted as done.",
        "Refreshed navigation so your repositories, insights, and goals are easier to reach.",
        "As always, the agent proposes and OpenThunder verifies: provider agents never compute the verdict, and nothing reaches done without evidence."
      ]
    },
    {
      "version": "0.6.6",
      "date": "2026-09-11",
      "title": "A plain-language review brief on every verdict",
      "notes": [
        "New: the Can I Ship? page now gives you a Review Brief, a clear, human summary of the verdict: what is actually proven, what still needs checking, and the single next useful step. It is honest by design (it tells you when a number is not test coverage and when evidence needs refreshing).",
        "One click copies the brief, with its version and evidence limits, so you can paste it straight into a pull request or hand it to a reviewer.",
        "Under the hood: verdict evidence now records the exact commit it was computed against, so an approval can never carry over to a later change by mistake."
      ]
    },
    {
      "version": "0.6.5",
      "date": "2026-09-05",
      "title": "Terminal fix: nothing hidden behind the safety bar",
      "notes": [
        "Fixed: the command safety bar at the bottom of the terminal (the Safe, Caution, or Danger verdict) no longer covers the last lines of output. The terminal now makes room for it, so you can always see your most recent commands."
      ]
    },
    {
      "version": "0.6.4",
      "date": "2026-09-04",
      "title": "Build with your own Codex or Claude Code login",
      "notes": [
        "New: run Missions with the official Codex agent using your own Codex login, no API key needed, right alongside Claude Code. Both are local agents: your source stays in your workspace, usage belongs to your account, and Codex threads resume across the build, review, and remediation steps.",
        "Connect OpenThunder to the tools you already use: openthunder mcp install wires it into Codex, Claude, or Cursor, and openthunder mcp doctor confirms the connection with a real handshake so you know it works.",
        "The Providers page now leads with local agents (Codex and Claude Code, using your login) before the cloud providers, and the fallback order prefers them.",
        "As always, the agent writes the change and OpenThunder proves whether you should ship it: provider agents never compute the verdict."
      ]
    },
    {
      "version": "0.6.3",
      "date": "2026-09-02",
      "title": "Preview vs Verified, and a verdict that keeps itself current",
      "notes": [
        "Honest proof levels: every verdict now says whether it is a Preview, Verified, or Approved. A Preview comes from static analysis (the change, coverage, architecture, and risk) without running the test suite; Verified means a ship-blocking claim was actually executed and is fresh against this commit; Approved adds a human sign-off bound to that exact commit. A Preview can never look like executed proof again, on the pull-request comment or the required check.",
        "New: openthunder watch, the return loop for the terminal. It keeps an eye on your repository and re-runs the ship verdict the moment a change is meaningful, a commit, a stage, an edit, a new file, and reprints the receipt in place. Keyless, local, and the exact same verdict can-i-ship gives you, just without having to ask again."
      ]
    },
    {
      "version": "0.6.1",
      "date": "2026-09-02",
      "title": "The Ship Room, now on every platform",
      "notes": [
        "Windows is current again: 0.6.1 ships signed on macOS, Linux, and Windows together, so every platform is on the latest Ship Room build.",
        "The pull request comment is now a full Ship Receipt: a large verdict, what this change did to shippability since the last commit, verified vs unproven claims with their provenance, evidence coverage and freshness, repository memory, and one-click links to the Proof Graph and Evidence Pack. It updates in place as the verdict improves."
      ]
    },
    {
      "version": "0.6.0",
      "date": "2026-08-20",
      "title": "The Ship Room: prove whether you should ship it",
      "notes": [
        "AI can write the change. OpenThunder now proves whether you should ship it. On the Can I Ship? page the verdict is live: it shows what share of the ship-blocking claims are backed by fresh, re-runnable evidence, ranks the single most valuable proof gap, and updates in place as you work.",
        "Close the next proof gap in one click: OpenThunder generates and verifies the missing test, then the verdict moves in front of you.",
        "Proof, not a mystery score: open the Proof panel to trace any claim to the source, command, and commit behind it, all backed by the tamper-evident evidence ledger.",
        "Recorder: replay how an AI change was produced, the edits and commands, grouped into sessions.",
        "Repository memory: OpenThunder remembers which files keep blocking your ships, and you can accept a reviewed risk. Acceptances are bound to the code and auto-reopen the moment that code changes, so nothing is silently suppressed.",
        "Rollout note: 0.6.0 is going to macOS and Linux first. The signed Windows build follows shortly; your current Windows version keeps working in the meantime."
      ]
    },
    {
      "version": "0.5.9",
      "date": "2026-08-14",
      "title": "A terminal that leaves evidence",
      "notes": [
        "New: Command Provenance. Every command you run in the terminal is tagged with a safe / caution / danger read and what it touches (files, network, packages, git history), and the session keeps an evidence log you can open and clear. It is deterministic, needs no AI, and never leaves your machine , so the terminal now records what happened instead of losing it to scrollback.",
        "Fixed: image files now preview correctly in the file viewer instead of showing a broken-image icon.",
        "Fixed: in a full-screen diagram, the \"Press Esc to go back\" hint no longer hides behind the window buttons."
      ]
    },
    {
      "version": "0.5.8",
      "date": "2026-08-11",
      "title": "Ask Codebase, sharper Can-I-Ship, and a richer Repo Brief",
      "notes": [
        "New: Ask Codebase (under Learn) answers questions about any repository with grounded, cited answers and diagrams, built from your real code graph and git history. Fund it with your own AI key, your Skills Tech Talk credits, or a Claude Code login you already have installed.",
        "Repo Brief now pairs an AI senior-engineer summary with the architecture diagrams, has clearer funding options, and can hand a repo off to Repo Mastery and Skills Tech Talk to practice defending it.",
        "Can I Ship? gives a sharper signal: per-file attention is more discriminating, and a missing test no longer forces a \"critical\" verdict on its own.",
        "Cloud onboarding copy is clearer, Can-I-Ship picks smarter defaults for what to compare against, and the Linux installers are now GPG-signed."
      ]
    },
    {
      "version": "0.5.7",
      "date": "2026-07-28",
      "title": "The Windows app is now signed",
      "notes": [
        "The Windows installer is now code-signed with Azure Trusted Signing, so Windows no longer shows the SmartScreen \"unknown publisher\" warning when you install or update. The signature verifies the app is genuinely from Skills Tech Talk, LLC.",
        "Windows users also catch up to everything since 0.5.4 in one update: AI Diagram, multi-language IntelliSense (Java, C#, Python, and more), the Design view, and test-coverage evidence."
      ]
    },
    {
      "version": "0.5.6",
      "date": "2026-07-27",
      "title": "Power AI Diagram with your Skills Tech Talk plan on desktop",
      "notes": [
        "AI Diagram's \"Skills Tech Talk credits\" option now works in the desktop app, not just on the web. If you're signed in with Skills Tech Talk, you can meter a diagram generation against your plan instead of using your own AI key.",
        "Signed out, nothing is sent anywhere, and the free deterministic system map still needs no AI at all, so the desktop app keeps its local-first, no-egress default."
      ]
    },
    {
      "version": "0.5.5",
      "date": "2026-07-27",
      "title": "Faster navigation and a more actionable Can-I-Ship",
      "notes": [
        "The command palette (Cmd-K / Ctrl-K) now finds every destination, including the pages tucked under Advanced. Type a name or a word like \"verdict\" or \"coverage\" and jump straight there.",
        "The Can I Ship? verdict now links each fix step to the page that helps you resolve it (Coverage, Security, or What Changed), so a HOLD is one click from the fix.",
        "AI Diagram moved under Learn, and you can now choose how to power it: your own AI key, Skills Tech Talk plan credits, or a free deterministic system map that needs no AI at all."
      ]
    },
    {
      "version": "0.5.4",
      "date": "2026-07-23",
      "title": "SVG files preview in the editor",
      "notes": [
        "Opening an SVG file now shows the rendered image instead of its XML source. Raster images (PNG, JPG, GIF, WebP, and more) already previewed; this closes the SVG gap."
      ]
    },
    {
      "version": "0.5.3",
      "date": "2026-07-23",
      "title": "Paste several images at once",
      "notes": [
        "You can now paste two or more images into the terminal composer in one go (for example, several image files copied together from Finder), not just a single image. Drag-and-drop of multiple images already worked and still does."
      ]
    },
    {
      "version": "0.5.2",
      "date": "2026-07-23",
      "title": "Design moves in with the other lenses",
      "notes": [
        "The Design view (UML diagrams, system-design doc, and candidate ADRs generated from your code) now lives in the Understand section, next to Architecture and the Deep Map, where the rest of the read-the-code lenses are."
      ]
    },
    {
      "version": "0.5.1",
      "date": "2026-07-22",
      "title": "OpenThunder closes the test-coverage loop",
      "notes": [
        "OpenThunder now runs your test coverage for real: it detects your project's coverage command (Vitest, Jest, pytest, dotnet, Maven or Gradle with JaCoCo), runs it behind the same trust gate as the terminal, and measures exactly how much of your change is covered.",
        "It writes the missing tests, then proves them: for a change with uncovered lines, OpenThunder drafts the unit tests, runs them, and keeps only the ones that pass and actually raise coverage, then flips the ship verdict from caution to ship. The generated tests are left as a labeled, uncommitted proposal for you to review; nothing is ever auto-committed or trusted on assertion.",
        "The coverage signal now gates everywhere it should: the Can I Ship verdict, the Change Pack, and the can_i_ship tool your agents call all downgrade a risky change that ships with no tests."
      ]
    },
    {
      "version": "0.5.0",
      "date": "2026-07-22",
      "title": "Design, side-by-side Thunder, test-coverage evidence, and Python",
      "notes": [
        "New Design view: OpenThunder reads your repository and generates UML diagrams (component, class, and sequence), a system-design doc, and candidate ADRs, all derived from the code and clearly marked for your review. See the design of what you actually built, not a stale wiki.",
        "See any file as ThunderLang, side by side: open a TypeScript, JavaScript, Python, Java, or C# file and toggle a pane that shows that file's equivalent ThunderLang interpretation, synced as you move between files.",
        "Test coverage is now first-class verification evidence: OpenThunder reads your coverage reports (lcov, Istanbul, Cobertura), computes how much of what you changed is actually covered, and flags changes that added no tests. A risky change that ships with no tests is now held or cautioned, not waved through.",
        "Python joins TypeScript, JavaScript, Java, and C# as a first-class language in the editor: open a .py file and get completions, real-type hovers, go-to-definition, and accurate errors, running locally with zero setup."
      ]
    },
    {
      "version": "0.4.9",
      "date": "2026-07-22",
      "title": "Thunder: the living spec OpenThunder writes from your code",
      "notes": [
        "New Thunder view: OpenThunder reads your repository and writes a readable ThunderLang spec of what the code actually does , capabilities, APIs, events, and data stores , so you can finally see the intent instead of guessing at it.",
        "It is verified, not just generated: each part of the spec is checked against the code and marked satisfied, drifted, or missing, so you know where reality and intent have diverged.",
        "Read, refine, and adopt: edit the generated spec in the editor and adopt it into your repo (intent/main.thunder) when it is right, turning a throwaway summary into a spec you own.",
        "Everything here is deterministic and keyless: the spec and its verification come straight from your code, with no AI key required and nothing uploaded."
      ]
    },
    {
      "version": "0.4.8",
      "date": "2026-07-22",
      "title": "A real editor for Java, C#, and ThunderLang , and a fixed dashboard",
      "notes": [
        "IntelliSense now covers Java and C#, not just TypeScript: open a Java or C# file and OpenThunder can run its language server locally (with one-click install of the Java/C# server the first time), giving you completions, real-type hovers, go-to-definition, and accurate errors.",
        "ThunderLang is now a first-class language in the editor: syntax highlighting, autocomplete, hover, and ready-to-use starter templates so you can begin writing intent in seconds.",
        "Fixed a crash that could take down the dashboard on load (a decision record was being rendered incorrectly); the whole app is hardened against it now.",
        "The workspace feels more like an IDE: Files and Repositories moved up top next to the Terminal.",
        "The Deep Map now groups every language into subsystems (packages/namespaces), so large repos in any language are easy to navigate."
      ]
    },
    {
      "version": "0.4.7",
      "date": "2026-07-21",
      "title": "Real IntelliSense in the editor",
      "notes": [
        "The editor now has project-wide IntelliSense for TypeScript and JavaScript: completions, hovers with real types, signature help, go-to-definition, and accurate error squiggles that resolve your imports, Node globals, and JSX , powered by a TypeScript language server that OpenThunder runs locally for the open repo.",
        "Fixed the phantom editor errors from the last build (spurious 'Cannot use JSX', 'Cannot find name process', and 'Cannot find module' warnings): those were a file-scoped checker with no project loaded. Real diagnostics now come from the language server.",
        "IntelliSense stays on your machine: the language server runs locally against the repo you opened; nothing is uploaded."
      ]
    },
    {
      "version": "0.4.6",
      "date": "2026-07-21",
      "title": "A real code editor, a deeper Deep Map, and more model providers",
      "notes": [
        "Files is now a real editor: opening a file gives you the full VS Code editor (Monaco) with syntax highlighting for 50+ languages, multi-cursor, find and replace with regex, a minimap, code folding, bracket-pair colors, and the command palette (F1). Cmd/Ctrl+S saves, and the AI Tab-completions and selection actions are kept.",
        "The Deep Map now reads a Java or C# repo end to end: every endpoint (with its auth), the entity diagram with relationships and cardinality, the layered request flows (controller to service to repository to table), and the codebase grouped into subsystems by package/namespace.",
        "Many more model providers: direct OpenAI and Anthropic, GitHub Models, Hugging Face, Qwen, Kimi, Z.AI, plus a Custom OpenAI-compatible / Anthropic-compatible endpoint and local runtimes (LM Studio, vLLM, llama.cpp).",
        "Large diffs are fast again (the diff view no longer renders every line at once), and a stray white bar next to the terminal is gone.",
        "The $49 Repo Snapshot is now a full report: an A to F ownership grade, code health, security and AI-security summaries, richer architecture, and a downloadable PDF."
      ]
    },
    {
      "version": "0.4.5",
      "date": "2026-07-20",
      "title": "Upgrades that never look like a crash",
      "notes": [
        "Fixed the macOS 'OpenThunder quit unexpectedly' report that could flash during an automatic update. The app now shuts its local engine down cleanly before restarting, so an upgrade is just an upgrade, not a scare.",
        "An upgrade is no longer mis-recorded as a crash internally, so a healthy version is never wrongly held back from auto-updating.",
        "This lands for every upgrade starting from this build; the very next update after installing 0.4.5 is clean."
      ]
    },
    {
      "version": "0.4.4",
      "date": "2026-07-20",
      "title": "Auto-update that just works, and the license shown only once",
      "notes": [
        "Auto-update is reliable again: the 'Restart to update' prompt is now always reachable. Previously, if the license screen appeared, the update banner was hidden behind it and a downloaded update could not be applied.",
        "The End User License Agreement is shown only once: acceptance is now saved in a durable location, so it no longer re-appears after a restart or an update.",
        "If a previous update got stuck, reinstalling the latest signed download once will carry you onto this build; auto-update takes over cleanly from here."
      ]
    },
    {
      "version": "0.4.3",
      "date": "2026-07-20",
      "title": "Java + .NET first-class, a Verified AI Change gate, and a signed release",
      "notes": [
        "Java and .NET are now first-class: Compile / Build / Run tests work for Maven, Gradle, and .NET, and can-i-ship's 'is this change actually tested?' signal recognizes JUnit and xUnit/NUnit tests, exactly like it does for npm , so a Java or C# change no longer looks untested.",
        "New: the Verified AI Change gate. Run 'openthunder verify-change' to put an AI-generated change through tests, security scanning, a ship/caution/hold verdict, and a required human approval before it is ship-eligible, all captured in a signed evidence pack. It can also block a merge in CI.",
        "Security: build and test commands no longer inherit your secrets (an env-scrub strips secret-shaped values before running), and internal error details no longer leak to callers.",
        "Signed releases: this build's supply-chain evidence is now Ed25519-signed. Verify this download yourself from the downloads page, or browse OpenThunder's analysis of famous open-source repos at openthunder.dev/proof.",
        "Onboarding no longer shows 'Cannot reach server' before a repo is connected, and 'Master this repo' is now clearly offered two ways: with AI in Skills Tech Talk, or without AI in Repo Mastery.",
        "Optional, off by default: anonymous usage counts to help improve OpenThunder (opt in with OPENTHUNDER_TELEMETRY=1; never sends code, paths, or repo names, and never runs in Local-Only mode)."
      ]
    },
    {
      "version": "0.4.2",
      "date": "2026-07-20",
      "title": "A proper first-run installer, and Run & Test that finds your tools",
      "notes": [
        "First-time installs now show the license agreement: opening the OpenThunder disk image presents the End User License Agreement with Agree or Disagree before it installs, like any signed app. Auto-updates are unaffected, so you are never re-prompted on upgrade.",
        "Run & Test finds your toolchain: Compile, Build, Lint, and Run tests now use your real shell PATH, so pnpm, node, and other tools resolve even when OpenThunder is opened from the Dock (no more 'command not found: pnpm').",
        "Practice & Defend works everywhere: the Skills Tech Talk hand-off no longer errors when there is no cloud workspace, so you can send a repo to practice from the desktop app.",
        "A more reliable mobile web app and assorted stability fixes."
      ]
    },
    {
      "version": "0.4.1",
      "date": "2026-07-19",
      "title": "On ThunderLang 0.4.1",
      "notes": [
        "Upgraded to ThunderLang 0.4.1: the language core behind Intent proofs, cross-language conformance, and Build & Debug is now on its latest release, with all the capabilities you already have.",
        "Improved OpenThunder Cloud sign-in reliability, so connecting this app to the cloud is smoother. The desktop app itself still needs no sign-in.",
        "Reliability and plumbing improvements across the board."
      ]
    },
    {
      "version": "0.4.0",
      "date": "2026-07-19",
      "title": "Missions you can trust: reliable runs, live progress, and edits on your repo",
      "notes": [
        "Run Mission now works end to end: retries, failures, and cancels always reach a clear result instead of getting stuck \"in progress\".",
        "Live progress: watch the agent read, edit, and run commands as it works.",
        "Cancel actually stops a running mission and its process instantly.",
        "Missions run against the project you choose: a new Project picker in New Mission points the builder at your repo, not a scratch folder.",
        "More responsive during runs; verification commands now support real shell syntax."
      ]
    },
    {
      "version": "0.3.1",
      "date": "2026-07-19",
      "title": "More room to work: terminal lanes, split panes, and integrated tooling",
      "notes": [
        "A roomier workspace: the terminal opens more lanes and split panes with integrated tooling, so you can run, watch, and edit side by side without switching away.",
        "Cleaner terminal layout: tabs now sit above the workspace controls, the tab and split-pane controls are tighter, and an empty split-pane column no longer shows up."
      ]
    },
    {
      "version": "0.3.0",
      "date": "2026-07-19",
      "title": "ThunderLang, end to end: build, test, and debug 11 languages, and prove your intent",
      "notes": [
        "Build & Debug, for real: a new tab runs Compile, Build, Lint, Run tests, Run, and Debug across all 11 languages OpenThunder understands (TypeScript, Python, Java, C#, Go, Rust, Ruby, PHP, Kotlin, Scala, Elixir). Commands run in your shell, locally, with live output and a pass/fail result, and nothing runs until you trust the exact command set.",
        "Prove your intent: OpenThunder now compiles authored ThunderLang and emits an honest proof, a guarantee is 'verified' only when a real test proves it, otherwise it is declared or needs verification, never asserted. Run it from the CLI (`openthunder intent prove`) so it drops straight into CI.",
        "Same intent, every language: cross-language conformance compiles a decision to TypeScript, Python, C#, and Java, runs the same cases against each, and grades the real outputs side by side, skipping any language whose toolchain you do not have installed.",
        "Faster where it was slow: the diff viewer no longer re-parses on every click, and the product-diff / brief / change views stay warm across restarts.",
        "Connect from another device: the Environment panel now shows the exact Tailscale address to reach OpenThunder (with a copy button), even before you expose it.",
        "Run Mission never fails in silence: if a run cannot start, it now tells you exactly why (a stuck mission, the wrong state, or an unreachable server) instead of doing nothing.",
        "Built on ThunderLang 0.3.0, now covering all 11 languages end to end: discover, lift, compile, and prove."
      ]
    },
    {
      "version": "0.2.7",
      "date": "2026-07-19",
      "title": "Sharper security scanning: fewer false alarms, no hidden real risks",
      "notes": [
        "The security scanner is calibrated to stop crying wolf: weak-hash used for cache keys, database `.exec` calls, and safe DDL statements no longer show up as high-severity findings, so the results you see are the ones worth acting on.",
        "Every quieted finding is demoted with an explanation, never silently dropped, and real vulnerabilities (command injection, SQL injection from request input, disabled TLS verification) still fire at full severity , verified by a new suite of precision and recall tests.",
        "Fixed a scanner bug where a real issue could be hidden if it appeared later in the same file as a detector's own example pattern."
      ]
    },
    {
      "version": "0.2.6",
      "date": "2026-07-19",
      "title": "Ten more AI providers, and missions that never fail in silence",
      "notes": [
        "Bring your own key for ten more providers: DeepSeek, Mistral, xAI Grok, Google Gemini, Cohere, Azure OpenAI, Fireworks, Cerebras, Perplexity, and NVIDIA NIM. Each becomes available the moment its API key is set, and works as a review and chat provider.",
        "Standard-mode tasks are no longer silent: when a task-looking prompt is sent as plain chat, the terminal now says so clearly and points you to the Verified shield, so a mission never appears to fail for no reason.",
        "A lost or slow mission stream now tells you what happened and how to check the outcome, instead of just stopping.",
        "The OpenThunder bolt mark now shows in the sign-in header, and Cloud sign-in reliably returns you to Cloud."
      ]
    },
    {
      "version": "0.2.5",
      "date": "2026-07-19",
      "title": "Drag-and-drop images, and a Verified/Standard terminal switch",
      "notes": [
        "Drag an image straight onto the terminal composer to attach it , the whole panel is now a drop zone with a clear \"Drop image to attach\" hint, not just the input line. Pasting still works too.",
        "New Verified/Standard switch in the terminal: Verified turns your tasks into governed missions with an evidence record; Standard sends them straight to the selected model, with no mission or evidence created.",
        "Clearer message when Practice can't embed Skills Tech Talk, so it points you to open it in a tab instead of showing a dead frame."
      ]
    },
    {
      "version": "0.2.4",
      "date": "2026-07-18",
      "title": "Run your services from OpenThunder, and AI implementations that find themselves",
      "notes": [
        "Trust and Run is live: review the exact commands OpenThunder detected for your project, trust them, then start and stop services with live health checks; nothing runs until you trust it, and it stays local.",
        "AI-managed implementations now discovers AI-written code regions directly from your repository (no manifest file needed), and you can adopt an existing range in-app and verify it here.",
        "The ecosystem switcher now uses ThunderLang official brand mark."
      ]
    },
    {
      "version": "0.2.3",
      "date": "2026-07-18",
      "title": "Faster navigation, terminals that survive reloads, and lossless upgrades",
      "notes": [
        "Navigation is noticeably snappier: pages cache their data and prefetch in the background, and the app bundle is about half the size.",
        "Terminals now survive a reload or an update: your shells reconnect with their scrollback intact, on both the web and desktop apps.",
        "Upgrades feel like nothing changed: your open files with unsaved edits, tabs, active page, and window position are restored after a relaunch.",
        "The ecosystem switcher shows product logos, and the terminal appearance panel was redesigned with a live preview and fit-to-space.",
        "Under the hood: describe how your project runs with Run Profile editing, tamper-evident signed Evidence Packs, and accessibility improvements."
      ]
    },
    {
      "version": "0.2.2",
      "date": "2026-07-18",
      "title": "Interface polish across the dashboard, status bar, and terminal",
      "notes": [
        "Cleaner dashboard: a titled Repository overview header and clearer Open missions labelling.",
        "The status bar and header carry hover tooltips and clearer states, including an active-provider Fallback label in place of a bare arrow.",
        "Clearer terminal messaging: when no terminal backend is available, OpenThunder now says exactly why (no local server, or the node-pty module missing) instead of one generic note."
      ]
    },
    {
      "version": "0.2.1",
      "date": "2026-07-18",
      "title": "Hop between the Skills Tech products from one place: a new ecosystem app-switcher",
      "notes": [
        "A new 9-dots app-switcher in the header lets you jump between all four Skills Tech products , OpenThunder, Skills Tech Talk, Skills Tech Network, and ThunderLang , on one account.",
        "In the desktop app the switcher opens OpenThunder Cloud (openthunder.ai) in your browser, so your local work and the hosted dashboard are one click apart."
      ]
    },
    {
      "version": "0.2.0",
      "date": "2026-07-18",
      "title": "Understand any codebase deeper: an entity-relationship data map, a richer Intent Atlas, and a better terminal",
      "notes": [
        "Deep Map is now an entity-relationship diagram: click any table to see its fields (SQL or NoSQL) with primary and foreign keys, and the relationships between tables. The data model is read straight from your schema, keyless, and no longer a flat list of names.",
        "The Intent Atlas goes deeper: every system expands to the routes it exposes, the data it touches, and the end-to-end flows that cross tiers, with each step traceable to the file it runs in, so you can actually see how the software works.",
        "The terminal gets a right-click menu (Copy, Paste, Select all, Clear) and a set of built-in color themes with full ANSI palettes: Tokyo Night, Catppuccin, Nord, Dracula, Gruvbox, One Dark, and more.",
        "Repository mastery hand-off: OpenThunder now packages an Intent Atlas, the data model, and a plain-language 'understand this repo like the engineer who wrote it' brief, ready to study in Skills Tech Talk. All safe-derived: names and structure only, never your source or secrets."
      ]
    },
    {
      "version": "0.1.14",
      "date": "2026-07-14",
      "title": "The terminal comes to the web, and a deeper Intent Scanner",
      "notes": [
        "Run a real terminal from the web app, not just the desktop: full shells over a local PTY, plus a Mission Router tab, as two separate entries whose tabs you can mix.",
        "Terminals now stay alive as you move between pages: your shells, scrollback, and running commands survive navigation instead of resetting.",
        "The Intent Scanner goes deeper: focused views for gaps, unverified claims, and contradictions; declared intent read from compiler artifacts; a security-sensitivity risk overlay; and it is now reachable over the API and MCP, and as a risk overlay on the Mission Atlas.",
        "Verify moved into the right-hand panel and now falls back to a branch-versus-base comparison so it stops coming up empty.",
        "Smaller, faster desktop build: the server bundle dropped from about 7 MB to 3.2 MB and release builds no longer stall."
      ]
    },
    {
      "version": "0.1.13",
      "date": "2026-07-12",
      "title": "A real multi-tab terminal, and understand any repo in one command",
      "notes": [
        "New multi-tab terminal: run shells and Claude Code side by side in named, draggable tabs, with SSH/Tailscale connections and a Ready panel that prepares the repo (toolchain, env) before you start.",
        "Discover Intent: OpenThunder now reports what a repo DOES , its product capabilities , from the code alone, keyless. Get a shareable Application Discovery Brief (what the system is, what changed, drift, coverage, a readiness verdict) and a change-scoped Change Pack (what a change touches + can-i-ship).",
        "Can-I-Ship is now intent-drift aware: a change that no longer matches the declared intent is flagged before you ship."
      ]
    },
    {
      "version": "0.1.12",
      "date": "2026-07-09",
      "title": "Intent verification: does your code still match its missions?",
      "notes": [
        "New Intent tab: OpenThunder checks your repo against its declared intent (from the SkillsTech Compiler) , API and event coverage, verified guarantees, unenforced never-rules, and drift where the code no longer matches, across every mission in the repo.",
        "Security: an optional 'Explain with AI' triage on each finding flags likely false positives and suggests the next step. Clearly labeled; findings work fully without a key.",
        "Reliability: fixed the ecosystem hand-off logos and a stale cloud-deploy path."
      ]
    },
    {
      "version": "0.1.11",
      "date": "2026-07-09",
      "title": "Cross-language architecture events and a cleanly notarized auto-update",
      "notes": [
        "Deep Architecture Map now traces domain events (publish/subscribe) across TypeScript, Python, Go, C#, and Java, and flags orphaned events: published but never consumed, or handled but never emitted.",
        "Intent: a new command reports where the implementation has drifted from the declared intent.",
        "A freshly Apple-notarized macOS build, delivered through our automated release pipeline, so every earlier version auto-updates cleanly with no Gatekeeper warning."
      ]
    },
    {
      "version": "0.1.10",
      "date": "2026-07-07",
      "title": "Smoother Missions, Control decisions, and a notarized rebuild",
      "notes": [
        "Fixed: Active and Recent Missions now scroll properly (the list could freeze).",
        "Practice embeds Skills Tech Talk reliably now, with an open-in-browser fallback if the embed is blocked.",
        "Control: approve a mission-contract verdict, or override it with a recorded justification, saved to the Evidence Ledger.",
        "New Decisions tab in Reports shows the recorded ship and override decisions.",
        "New: a heads-up when a mission goal is too vague to verify, before a wasted run.",
        "Signed and notarized for macOS."
      ]
    },
    {
      "version": "0.1.9",
      "date": "2026-07-04",
      "title": "Works embedded inside VS Code",
      "notes": [
        "Install the OpenThunder VS Code extension and the dashboard now opens right inside your editor, beside your code. No need to switch to a separate window.",
        "Open any view (Architecture, Security, Repositories, Trust, and more) as a tab in VS Code."
      ]
    },
    {
      "version": "0.1.8",
      "date": "2026-07-04",
      "title": "Trust & Privacy: see and control exactly what leaves your machine",
      "notes": [
        "New Trust & Privacy page: turn on Local Only to block all network access, and watch a live Data Activity log of exactly what leaves your machine (nothing, in Local Only).",
        "Your secrets stay yours: .env files, keys, and credentials are now withheld, and secret-shaped values redacted, before any code is sent to an AI provider.",
        "Fixed: creating a mission no longer fails when no AI provider is configured.",
        "Faster New Mission dialog."
      ]
    },
    {
      "version": "0.1.7",
      "date": "2026-07-04",
      "title": "Claude Code shows up as a builder, and a bigger app icon",
      "notes": [
        "Fixed: Claude Code (my login) now appears in the Builder and Reviewer menus in the installed app. It was there in 0.1.6 but the app could not find your installed claude command; now it does.",
        "Bigger, clearer app icon in the Dock and app switcher.",
        "This version installs itself automatically over 0.1.6."
      ]
    },
    {
      "version": "0.1.6",
      "date": "2026-07-04",
      "title": "Use your own Claude Code, and a properly signed Mac build",
      "notes": [
        "New builder and reviewer option: Claude Code (my login). Missions can now run on your own installed Claude Code, no API key needed; usage is on your own Claude account.",
        "The Mac app is now fully signed and notarized, so it opens without any Gatekeeper warning.",
        "Auto-update now works end to end on Mac (it needed the signed build)."
      ]
    },
    {
      "version": "0.1.5",
      "date": "2026-07-04",
      "title": "Claude Code in the terminal, Can-I-Ship verification, and shareable proof",
      "notes": [
        "Run Claude Code right inside OpenThunder: the Terminal now launches Claude and logs you in.",
        "Can-I-Ship: one click tells you ship, caution, or hold, with the exact reasons and file:line evidence.",
        "Blast radius on every change: see what an edit actually affects before you merge.",
        "Evidence Ledger: record a verified change as tamper-evident proof, then share it as a public link.",
        "Policy-as-code: enforce team rules (like 'auth changes need human review') on the verdict.",
        "Agent Review: see what the AI changed and whether it was verified.",
        "Smaller, faster app."
      ]
    },
    {
      "version": "0.1.3",
      "date": "2026-06-30",
      "title": "Seamless training, polished light theme, and Windows fixes",
      "notes": [
        "Train on this repo now works in one click, no setup.",
        "Polished light theme (no more washed-out screens).",
        "Fixed scrolling on Pricing and Prompts.",
        "No more Windows location permission prompt.",
        "More reliable startup with built-in diagnostics."
      ]
    },
    {
      "version": "0.1.2",
      "date": "2026-06-30",
      "title": "Colorful diagrams, full-screen view, and smoother scrolling",
      "notes": [
        "Architecture and Mastery diagrams are now vibrantly color-coded by component type.",
        "New full-screen button on every diagram (press Esc to exit).",
        "Fixed scrolling in the Architecture Lens when opened from a repository.",
        "Every build now shows its version and build number in the sidebar.",
        "Updates install in the background, so you never have to re-download.",
        "Bigger, better-centered app icon."
      ]
    },
    {
      "version": "0.1.1",
      "date": "2026-06-30",
      "title": "Faster start, prettier lenses, and one-click updates",
      "notes": [
        "Fixed a startup crash on some machines so the app always opens.",
        "The app now picks a free port automatically and always starts, even if another app is using the default one.",
        "Architecture and Security Lens redesigned: a circular score gauge, cleaner sub-scores, and polished critical-flow cards.",
        "Mobile-friendly lenses that adapt to small screens.",
        "Repository Mastery is richer: 10 diagrams, a repo mindmap, a large flashcard deck with answers, and an interview brief.",
        "New welcome flow: start locally in one click, no account or AI key required.",
        "In-app updates: see what's new and upgrade with one click."
      ]
    },
    {
      "version": "0.1.0",
      "date": "2026-06-29",
      "title": "First public build",
      "notes": [
        "Architecture Lens: open any repo and map its structure, flows, and scorecard.",
        "Security Lens: findings, STRIDE threat model, and risk checks before you ship.",
        "Verified changes: builder plus independent reviewer, with an evidence report.",
        "Repository Mastery: export a pack to Skills Tech Talk and train on the codebase.",
        "Local-first: analysis runs on your machine; cloud features are opt-in.",
        "Automatic updates: future versions install seamlessly, no reinstall."
      ]
    }
  ]
}