SkillsTech release notes

Ownership stops being a set of commands you run and becomes something you can see and feel while you work, across thirteen languages. Everything here is offline and private; your source is never uploaded.

A Mission Center that feels like an ownership IDE

• The Mission Center is now the home, not a raw file. A Trust Bar across the top always shows the repo, branch, local-only mode, "no source sent externally", your risk grade, proof state, and whether AI is on. The Ownership Loop becomes a guided eight-step stepper (Understand, Plan, Context, Apply, Review, Verify, Explain, Prove) with a live status on each step and the single best next action.
• Proof is a product, not a JSON file. A Proof Packet Preview shows the quality-gate checks, the review checklist, and a ready-to-paste PR description, with one-click Copy, Export, and Send to Workspace. Raw JSON is tucked behind Advanced.
• A Verification Console with tabs (Summary, Risks, Tests, Architecture, Secrets, Proof, Logs) surfaces the structured result first and raw logs last. A Can I Ship? panel gives a plain verdict, the reasons, and the human steps that remain before merge.
• An actionable Repo Map (framework, language, confidence, first files, ownership signals) and a first-class Intent card (missions, guarantees, never-rules) sit alongside a contextual assistant whose suggestions change with what you are looking at.
• Everything runs locally and is honest about what it does not know ("No test run detected", "Framework unrecognized") instead of guessing. Status is shown with clear color and text, never decorative icons.

Ownership, always on

• Ownership CodeLens. Every source file now shows, inline at the top: how many files depend on it (its blast radius), its framework risk, and a one-click AI context pack.
• Import hovers. Hover any import to see the target file's blast radius and risk, with a link to jump to it.
• A live guardrail in Problems. Save a changed file that is framework-risky or has a wide blast radius and SkillsTech flags it, "review and prove this change", before it ships. Scoped to files you have actually changed, so it guides instead of nagging. Tune it with skillstech.diagnostics.blastRadiusThreshold.
• A live Ownership panel. The SkillsTech sidebar now shows where you are in the loop (Understand, Review, Verify, Prove, Share), your proof count, and the single best next action, updating as you create and push proof. A Verified badge appears once you have proof that passed the Verify gate, showing your best verdict (clear / caution / high-risk), and completes the Verify stage.
• A status-bar glance. Your proof count (or an invitation to prove your first) sits in the status bar, one click from the panel.
• Verify is now a real gate. Verify combines an offline ownership gate, the framework risk and blast radius of every file you changed, with a clear/caution/high-risk verdict, together with the engine's risk, architecture, and security scan. The ownership gate always runs, even with the analysis engine off.
• Proof records the verdict. When you Prove a change, the review packet now captures the ownership-gate verdict (clear / caution / high-risk) and its counts, so your proof shows not just that you reviewed a change but at what risk level you owned it. Safe-derived: the verdict and counts carry no source or file paths.
• Symbol Callers and Callees. Two new commands work one level below the file graph. Right-click a function name to see where it is declared and every file that calls it, ranked by call count (a symbol-level blast radius); or run Symbol Callees on a selection or file to see which in-repo functions it calls and where those are defined. Fast name-based reference search across all supported languages.
• Export Ownership Report. One command (also a button on the SkillsTech panel) writes a shareable report: an ownership health score (0-100 with an A-F grade and a transparent factor breakdown, understanding, change proof, verification, risk control, portfolio backup) that tracks a trend across exports (up or down since last time), your stack, the highest-impact files by blast radius with their risk, your proof status, and an ownership posture. Exports both ownership-report.md and a self-contained, styled ownership-report.html (dark-mode aware, no external assets) you can open in a browser or share. Offline and safe-derived (paths and counts, never source). The report now also shows hotspot coverage: which of your highest-impact files have a proven change and which are still unowned (Prove records the changed file paths, so the report can mark them).
• Each surface has its own toggle (skillstech.codeLens.enabled, .hover.enabled, .diagnostics.enabled, .statusBar.enabled).

Write intent, get solid code (IntentLang)

• IntentLang comes to Studio. Open a .intent file and Studio runs the deterministic SkillsTech Compiler as you type, offline, no AI. You describe a mission (its goal, inputs, outputs, guarantees, and what it must never do); Studio checks it and, when you are ready, turns it into real code.
• Teaching diagnostics inline. The compiler catches intent that cannot hold, right in the editor: a duplicate-prevention guarantee with no idempotency signal, money without a currency, a guarantee no test covers, a "never" with nothing enforcing it. Each problem carries a plain-language fix.
• Transpile to any of twelve targets. One command turns a mission into TypeScript, Python, C#, Go, Java, Rust, SQL, Terraform, Bicep, Kubernetes, GitHub Actions, or Pulumi, each with a failing-by-default test so the work you still owe is visible. Money is always decimal, never a float.
• An honest proof artifact. Write Proof emits <Mission>.intent-proof.json: deterministic, source-hashed, and never marked "proven" until something actually verifies it.
• IntelliSense for .intent. Context-aware completions (block keywords, semantic types after a colon, target names on the target line, plus full-mission and examples snippets), hovers that explain each keyword and semantic type, and an Outline of every mission and its fields. And a first: hover a symbol in any generated file (TypeScript, Python, and the rest) and, if it came from a mission, you see that mission's intent, its goal, guarantees, and never-rules, right there in the code.
• Toggle it all with skillstech.intent.enabled; silence just the live diagnostics with skillstech.intent.diagnostics.enabled. Fully offline; your source never leaves your machine.

Coming soon

• SkillsTech Certified groundwork. Export Certification Proof (off by default) wraps your latest review packet into a safe-derived certification-evidence artifact for the SkillsTech Certified AI-Era Software Engineer Associate track: the Verify verdict, blast-radius counts, and a safe summary, never source. Sending to Skills Tech Certified is feature-flagged until the endpoint is live.
• Curriculum update suggestions. Studio can propose curriculum updates to Skills Tech Certified for the ownership features it ships (which certification domain and skill each affects, with suggested lesson, quiz, and mission notes). Studio only suggests; Skills Tech Certified authors the curriculum. Feature-flagged.

More languages

• Go, Rust, Ruby, PHP, C/C++, Swift, Scala, and Elixir join TypeScript/JavaScript, Python, C#, Java, and Kotlin, and Kotlin now has its own dedicated analysis (its own type-index import resolution, no longer sharing the Java pack). All the ownership signals above, plus Understand, Review change-impact, and Blast Radius, now work across all thirteen, with framework-aware risk (Gin/Echo, Actix/Axum, Rails/Sinatra, Laravel/Symfony, SwiftUI/Vapor, Ktor/Compose, Play/http4s/Spark, Phoenix/Ecto, unsafe-C calls, and more) and a real cross-file dependency graph, including C/C++ header blast radius and Swift/Kotlin/Scala/Elixir type-reference resolution.

Design and intent, for every role

Studio grows from an engineer's tool into a place where product, design, and engineering can all describe what a product should do, and agree on it, over one shared model. Every surface below is offline, deterministic, and honest about the line between what you author here and what another system decides: SkillsTech authors and captures; IntentLang validates; OpenThunder verifies; SkillsTech Workspace records approvals. Nothing here claims something is validated, verified, or approved on its own.

• Product Mission Builder. Author a product mission the way you would think it through: the problem, the evidence behind it (each item honestly classified, and an "observed" fact must carry a real source, the builder never assumes one), the outcomes and their metrics, scope, non-goals, the "never" rules, open questions, and who must approve. It saves the same mission the Product view reads, so what you author shows up where the team looks.
• Experience Contract Builder. Describe the experience: journeys, screens and their states, interactions, and content. Accessibility and responsive assertions are always recorded as claims, never as "verified", because a claim is proven by OpenThunder evidence and human review, not by writing it down.
• Journey Builder. Map a user journey as an ordered path of steps, each with its action, the screen it lands on, the touchpoint, and any friction. A friction note is a hypothesis by default, never silently marked as an observed fact.
• Conflict-Resolution Workspace. When two roles declare requirements that cannot both hold, the workspace shows the smallest conflicting set, the resolution options, and who must agree, and lets a human choose a resolution and write it back, without touching any raw internals. A conflict is resolved only when every required role approves.
• Semantic Diff. Compare two versions of an intent and see the impact grouped by role (product, experience, security, and the rest), the risks a change introduces, and a preview of which approvals it would invalidate. It is a structural diff of the intent, not an implementation verdict, and the approval preview is exactly that: a preview, until Workspace records the real invalidation.
• Approval Workflow. Version-specific approvals: an approval is bound to the version it approved, so when the intent moves forward, prior approvals go stale and need re-approval. AI can never grant an approval, the requester can never approve their own change, and every decision is captured in Studio pending Workspace, which remains the system of record.

Style Suggestions

• A design system you can choose, adapt, and keep. A Style Suggestions panel ranks polished style packs for your product, deterministically and offline (no AI, no network), with a plain-language reason for each. Preview a pack without changing anything, see a semantic Style Diff before you apply, and edit colors, typography, spacing, shape, and density by hand. Locked brand tokens and the approved logo cannot change without permission, so a style can never quietly replace your brand. Export to CSS variables or DTCG tokens. Accessibility is asserted from each pack, and verified by OpenThunder's Style Lens, never claimed here. AI-generated variations are a later, opt-in step, never required.

Role-friendly Studio

• Guided Engineering Mode. A first-class experience for newer contributors, not the engineering view with features removed. A Guided Contribution walks fifteen steps from understanding the customer problem to recording verified proof, always showing the current step, why it matters, the relevant artifacts, and what remains. It never fakes progress: a step that needs a real signal (a test run, an OpenThunder verification, a review request, an approval) cannot be completed without it, so proof is never recorded before approval. The advanced and technical detail is always one click away, and graduating to full Engineering Mode keeps your context.
• Explain and Teach, anywhere. For any artifact (a mission, a requirement, a journey step, an IntentLang construct, an OpenThunder finding, and more), a one-command explanation built deterministically from the intent graph, at the depth you want, in friendly, standard, or technical wording that never changes the meaning. The source intent and the advanced explanation are always reachable. An AI-enhanced explanation is available, clearly marked and optional; the deterministic one always stands on its own.
• Role-friendly review on the web. At app.skillstech.dev, the review console now adapts its wording to your role and preferred terminology (the same artifact, presented for a product manager or an engineer, with the technical term always one hover away), and a Review tab lets a reviewer approve, request changes, or reject a proposed change without needing any version-control knowledge. Decisions are captured pending Workspace, with the same honesty as the desktop: AI never approves, the requester cannot approve, and stale approvals need renewing.

A new look

• The shield. SkillsTech Studio, the web app, and the site now carry the SkillsTech shield mark and a refined indigo palette shared with SkillsTech Certified: a crisp, scalable vector logo and a calmer, more premium dark theme. Status is still shown with clear color and text, never decorative icons.

Reliability and correctness hardening (from an internal code review), plus one small addition. Ships with the next build.

• View My Proof. A new command lists all the proof you have created (kind, summary, date, and whether it is in your Proof Portfolio or local only), so the loop is complete: create, view, push.
• No more wedged commands. Every engine, git, and local-server call now has a timeout, so a hung analysis engine or unresponsive local server can no longer freeze Understand / Review / Verify / Prove.
• Large-file safety. The import graph skips very large generated or minified files, so Review and change impact stay fast on real-world repositories.
• More accurate analysis. A project is only labeled Next.js when there is a real Next.js signal (no more false framework labels in the overview or in proof). Hidden folders (.git, .github, .idea, ...) no longer inflate the file count.
• Smarter update checks. Offline update checks no longer retry on every launch, and you are only prompted when a genuinely newer version is available.
• Checkout unlock reliability (behind the dormant Pro flag): the post-purchase entitlement read now parses the account response correctly and works with either sign-in method, so a completed upgrade is reflected in the app.

Focus of this release: get a new user to real, credible proof in their first session, keep everything private by default, and make the whole experience clearer.

New

• Create Proof of Understanding. Turn any repository into a credible ownership artifact in one step, fully offline, no account, no changes required. It captures a real understanding baseline (project type, languages, entry points, structure) so you have proof of ownership from your very first session.
• Prove now always produces something. If there are no uncommitted changes to review, Create Proof of Ownership saves an understanding snapshot instead of an empty packet. Make a change and run it again for a full review packet.
• Push proof to your Proof Portfolio (opt-in). When you connect an account, you can push your local proof to your portfolio. It is always your choice; "Keep local" is always offered, and nothing is ever sent automatically.
• Pro plan (coming soon). SkillsTech Pro will unlock hosted analysis, more repositories, AI missions, and communication practice. Pricing and checkout arrive in an upcoming release.

Privacy and trust

• Private by default, explicit consent to share. Publishing proof to your public profile now asks for clear confirmation first, and sensitive material can never be shared more widely than its classification allows. Your proof stays private unless you decide otherwise.
• Your source code is never uploaded and never used to train AI. The Understand and Prove features run locally on your machine.
• Anonymous, opt-out product analytics. To improve the product we record anonymous usage events only: no source code, no file names, no personal data, just an opaque device id. It also respects your editor's global telemetry setting, and you can turn it off with skillstech.telemetry.enabled.

Improved

• Clearer, tool-agnostic language across the ownership loop (Understand, Review, Verify, Prove).
• Icons instead of emoji throughout the interface for a cleaner, more consistent look.
• Groundwork for one shared SkillsTech sign-in across products, rolling out soon (off by default in this build).
• Built-in update checks so you are always one click from the latest signed, notarized build.

Install and upgrade

• New install: https://skillstech.dev/download/mac
• Existing users: the app checks for updates automatically, or run "SkillsTech: Check for Updates".

The app is signed and notarized for macOS. Editing is always free and never requires sign-in.