Intent-ranked paths
Repository files are scored against the request so the most relevant code surfaces first.
AI coding workflow for VS Code
Connect AI reasoning to local tools and verifiable results.
WebCodingAgent gives an eligible AI service focused repository context, turns its responses into developer-controlled local actions, and verifies completed work inside VS Code.
You supply eligible AI access · Provider terms and limits remain in force
01 / Context
Help the model understand your repository.
WebCodingAgent ranks paths from your task, active editor, manifests, diagnostics, symbols, open files, and repository instructions. The selected AI receives focused working context instead of a blind prompt or repository dump.
Symbol-aware discovery
Language symbols and diagnostics connect the request to implementations, callers, and failures.
Instruction hierarchy
Repository guidance is discovered by scope and carried into every affected change.
Bounded model context
Absolute workspace paths stay local and only task-relevant context is sent to the AI service you select.
Live steering
Add direction while the task runs without discarding completed or verified work.
Continuity
Restart VS Code and reconstruct the task from its canonical event log.
02 / Execution
The AI reasons. Your editor acts.
An eligible AI connection supplies model responses. Commands and repository mutations execute through the desktop extension, where workspace trust, local approval, path confinement, cancellation, and redaction remain authoritative.
Connect
Select an AI service and connection method you are authorized to use.
Plan
WebCodingAgent combines its response with repository context, acceptance criteria, and risk.
Execute
Approved local tools run inside the repository boundary with conflict-safe file writes.
Verify
Tests, builds, diagnostics, and diffs decide whether the task is actually complete.
03 / Proof
Completion is a gate, not a feeling.
WebCodingAgent records what changed and which checks actually ran. High-risk work can bring in isolated security, QA, or architecture specialists who inspect their own evidence before clearing completion.
Tests and builds
Required validation commands must run successfully; a confident summary cannot substitute for output.
Diffs and hashes
Existing files change only against the content hash returned by an earlier read, avoiding silent conflicts.
Independent review
Specialists challenge risky changes from separate context rather than endorsing the primary agent by default.
Tamper-evident audit
Policy decisions and task events form a visible, hash-linked history for later review.
Repository-aware by design
Bring AI into your VS Code workflow.
Install wcagent, connect eligible AI access, and keep repository actions under your local approval settings.
