Alurkode

Responsible coding intelligence for programming education.

Alurkode records assignment-scoped coding provenance and provides explainable clean-code and resource-aware learning guidance during an explicitly started session. It connects regular VS Code to a role-aware Alurkode course without turning a behavioral signal into an unsupported accusation.

Before you install

• Alurkode requires VS Code 1.132 or newer.
• You need an Alurkode account and enrollment supplied by your instructor or institution.
• The Proska deployment uses https://alurkode-api.proska.id; another institution may supply a different HTTPS API address.
• Read the privacy notice, terms of use, and support guide.

Start a student session

1. Install Alurkode from the VS Code Marketplace in regular VS Code. Extension Development Host is only for extension developers.
2. Open the assignment folder.
3. Run Alurkode: Sign In using the account activated from the course invitation.
4. Confirm the idle status bar displays Alurkode: @username; select it to inspect your account identity if needed.
5. Run Alurkode: Start Coding Session, choose an open enrolled assignment, and read the recording notice.
6. Confirm the red Alurkode recording indicator is visible while working.
7. Run Alurkode: Stop Coding Session. Recording stops locally immediately; wait for synchronization when connectivity is available.

Installing Alurkode, opening VS Code, signing in, or opening the web dashboard does not begin recording. If a student does not explicitly start a session and accept the notice, coding is not recorded and the dashboard can only report that no session was started.

If connectivity fails, Alurkode keeps pending events and finalization details in encrypted extension storage and retries automatically. The original stop time is retained, so offline time after stopping is not counted as coding. Use Alurkode: Retry Pending Session Sync if needed. Do not uninstall the extension or clear its storage before synchronization completes.

What students receive

For JavaScript, JSX, TypeScript, TSX, Python, Java, and the bounded C subset, Alurkode can analyze the current document locally. C++, C#, Go, Rust, PHP, and Kotlin receive validated lexical line classification plus language-labelled provenance and VS Code-supplied diagnostics, but no v0.13 deep semantic/environmental/project analyzer. Depending on assignment policy, guidance appears live, after the session, or only in the instructor report. Use Alurkode: Analyze Current File, Alurkode: Analyze Project, or Alurkode: Show Learning Feedback to open it manually. Project analysis is intentionally bounded to at most 50 registered files and models static imports/includes, calls, cycles, duplication, reachability candidates, complexity, coupling/cohesion proxies, error handling, and mutation where each language capability permits. Supported, partial, unresolved, and unavailable states remain visible; this is not proof of whole-program correctness or complete runtime architecture.

Clean-code recommendations provide:

• the measured pattern and configured threshold;
• why the pattern matters;
• a concrete action and learning objective;
• generic before/after examples;
• confidence, effort, validation guidance, and trade-offs.

Resource-aware recommendations identify static patterns that may increase CPU, memory, network, or I/O demand. Static source analysis does not measure electricity and never invents kWh or CO₂ values.

Phase 13 counts editor rows, code-containing source rows, comment-only rows, mixed code/comment rows, and blank rows separately. Mixed rows are source rows. Multiline/raw literal and nested-comment boundaries follow the registered language scanner. C/C++ headers are selected by VS Code's reported language ID; filename extension alone is insufficient.

Semantic and environmental events normally send source hashes, metrics, rule evidence, and recommendations—not raw source text. Function and class identifiers remain excluded unless policy explicitly enables them. Source snapshots require both instructor policy and local student opt-in.

Project analysis is local-first and uses one assignment-controlled upload mode: local only, aggregate totals/capabilities, session-scoped hashed relationships, or institutionally approved relative named relationships. Raw source and code excerpts are excluded from every mode. The selected mode is disclosed before recording starts; opening project analysis while idle does not create session telemetry.

AI-assistance evidence

When permitted, Alurkode: Declare AI Assistance lets a student voluntarily record the assistant and whether the help involved generated code, autocomplete, chat, debugging, explanation, or another category. Prompts and responses are not collected.

An assignment may also permit recognized-extension context. An extension being installed or active does not prove that it generated or changed code. Bulk insertion and focus patterns are behavioral context and cannot identify ChatGPT or another external application.

Alurkode keeps these categories separate:

• Direct declaration: what the student voluntarily reports.
• Cooperative direct evidence: a verified receipt from an approved connector; this is currently Not available.
• Contextual evidence: relevant extension state or environment information.
• Behavioral signal: an editor pattern such as a bulk insertion.
• Unknown source: what the available evidence cannot establish.

None is an automatic academic-misconduct verdict.

Errors, tests, coverage, and benchmarks

Assignment policy may permit compiler/linter diagnostic snapshots, task lifecycle, debug lifecycle, sanitized terminal-command metadata, and observable execution outcomes. Alurkode does not capture terminal stdout or stderr. Success or failure is recorded only when VS Code exposes an exit code; debug completion remains unknown when the public API does not expose an outcome.

Alurkode: Import Test Coverage Summary supports Istanbul/Jest coverage-summary.json, coverage.py coverage.json, and JaCoCo jacoco.xml. The student chooses and confirms the report. Alurkode sends normalized totals and a SHA-256 report hash—not source code, file-level coverage, or the report itself.

Alurkode: Run Controlled Benchmark uses a trusted .alurkode/benchmark.json, an executable allowlist, no shell, bounded repetitions, a timeout, a correctness oracle, and an environment fingerprint. It records repeated child-process wall-clock durations and uncertainty. It does not directly measure CPU time, peak memory, electricity, kWh, or CO₂. If—and only if—the assignment enables controlled carbon estimation and the benchmark declares institution-approved average power, power uncertainty, system boundary, grid factor, source, region, and observation date, Alurkode may calculate a low/central/high operational estimate. That output is labelled modelled, excludes embodied/out-of-boundary impacts, and is not suitable for cross-student grading.

Privacy boundary

During an explicitly active session, assignment policy may permit editor/file/focus metadata, sanitized command metadata, diagnostics, observable execution outcomes, hashes or snapshots, semantic/resource findings, recognized-extension context, voluntary AI declarations, coverage totals, and controlled timing results.

Alurkode does not record:

• browser history or visited URLs;
• screenshots or screen recordings;
• clipboard contents or raw keylogging;
• terminal output;
• AI prompts or responses;
• contents of other applications.

Students can use the web dashboard to see their own reports, consent receipts, contextual notes, and data-rights requests. Instructors, reviewers, and administrators receive only role-authorized views.

Commands

Support

See the support guide or email alurkode@proska.id. Never send passwords, access tokens, service credentials, raw student source code, or unredacted student records in a support request.

Alurkode is an evaluation and research prototype. A real classroom deployment requires appropriate institutional privacy, security, ethics, accessibility, and assessment review.