Skip to content
← back to projects
2026·shipped·Forward Deployed Engineer

AI-Native IDE: Intent-Code Traceability

A VS Code extension that links business intent, code AST, and agent action through content-hashed sidecar traces so parallel agents can share context.

10 Academy TRP · 1-week brief
TypeScriptVS Code APIASTYAML

Architecture

VS Code extension
Editor-native intent + trace layer
↓ calls out to
active_intents.yaml
Business intent
agent_trace.jsonl
Tool call log
Deterministic hooks
Enforce rules
Content hashing
Parallel agent sync

Highlights

  • ▸Sidecar files (active_intents.yaml, agent_trace.jsonl) capture the missing layer between prompt and diff.
  • ▸Deterministic hooks enforce architectural rules before an agent ever writes to disk.
  • ▸Content-hashed traces let parallel agents share context without a central coordinator.

The gap this fills

When agents modify code, the diff is only half the story. The other half is the intent behind the edit and the trace of tools the agent used to get there. This project makes both explicit as first-class artifacts inside the editor.

Artifacts

  • active_intents.yaml: the business intent currently in play, versioned alongside the code.
  • agent_trace.jsonl: an append-only log of tool calls, keyed by content hash.
  • Deterministic hooks: rule engines that block violating diffs before they land.

Parallel agents

Multiple agents can operate on the same repo without a central lock because their traces are content-hashed. Two agents converging on the same change produce identical hash prefixes and can merge; divergent work is flagged for human review.