hAIve — VS Code Extension

Surface team memories, gotchas, and architectural decisions inline while you code.
Never let your AI agent forget what your team knows.

Getting started

1. Install the extension (from the Marketplace, or a local .vsix — see Installation).
2. Initialize hAIve in your repo — open the Command Palette and run hAIve: Initialize in This Workspace (or haive init in a terminal). This creates .ai/memories/.
3. Open any file — anchored team knowledge appears inline (CodeLens at the top of the file), in the hAIve sidebar (Activity Bar), and in the status bar. Press Ctrl/Cmd+Shift+B to get a briefing for the current file, or Ctrl/Cmd+Shift+H to search memories.

The extension reads .ai/memories/** directly, so it works even without the haive CLI on your PATH. Cockpit/eval/sensor/doctor features call the CLI — set haive.cliPath if haive isn't on PATH.

Features

Strategic Cockpit

The Strategic Cockpit turns hAIve's CLI observability into a one-shot VS Code view:

• doctor scores: protection, context quality, corpus quality, harness coverage
• eval quality: score, recall, precision, MRR, sensor catch-rate, baseline delta
• corpus health: active/retired memories, stale, pending, anchorless, prune candidates
• memory impact: high/medium/low tiers and top records by demonstrated utility
• sensors: warn/block/autogen/fired counts and sensor records
• tool usage: top hAIve tool calls from the usage log
• strategic risks: noisy retrieval, unfired sensors, stale/pending/prune work

Use hAIve: Refresh Observability to rebuild the cockpit from the official CLI JSON commands (doctor, dashboard, eval, stats, memory impact, sensors list, memory lint).

Discipline Inbox

The Discipline Inbox is the human review queue for keeping context healthy:

• proposed/draft/stale/lint findings that need review
• prune candidates and deprecated records
• high-read memories that have never been marked applied
• autogenerated or warn-only sensors that need human promotion/rejection
• eval misses when retrieval or sensors regress

Right-click items to open the memory, mark it Applied or Rejected, run a suggested fix, approve/reject memories, or promote a vetted sensor to block.

Harness Health

The Harness Health view surfaces haive doctor scores — protection, context quality, corpus quality, and harness coverage — so you can see at a glance whether the agent harness is healthy before handing work to an agent. Use hAIve: Run Health Check to (re)compute the scores, and hAIve: Sync Memories to pull the latest shared corpus.

🧠 Inline CodeLens

Files with anchored memories show a count at the top:

🧠 hAIve: 3 memories — ⚠️ 1 gotcha · 🏗 1 architecture · 📐 1 convention
  ⚠️  pg Pool — must set max:5 in production
  🏗  DB — always call migrate() at startup
  📐  UUID as PK — never use sequential integers

Click any lens to jump to the memory file.

📋 Activity bar sidebar

Browse memories from the dedicated hAIve icon in the Activity Bar (left strip). Memories are grouped by type; when you edit a file, the tree prioritizes anchors tied to that path.

Older builds placed the panel under Explorer — the activity bar placement keeps institutional knowledge one click away.

Example tree:

⚠️ Action Required (2)
  └─ Breaking change: DELETE /users/:id removed
  └─ express ^4 → ^5 major bump
📄 This File (db.ts) (2)
  └─ DB migrations on startup
  └─ UUID primary keys
🏗 Architecture (2)
📐 Conventions (3)
🎯 Decisions (1)
⚠️ Gotchas (4)

🔴 Status Bar

Bottom status bar shows total memories and flags action_required:

⚠️ hAIve: 9 memories · 2 action required

⚡ Auto-reload

The extension watches .ai/memories/** and refreshes automatically when you run haive memory save, haive sync, or edit memory files directly.

Human context routines

The extension is designed to reinforce a small daily discipline:

1. Run Refresh Observability at the start of a work block.
2. Clear the Discipline Inbox before handing work to an agent.
3. Use Get Briefing for This File before editing policy-heavy files.
4. Mark surfaced memories Applied when they changed your decision.
5. Mark noisy/outdated memories Rejected so impact scoring can demote them.
6. Promote only vetted sensors to block; leave uncertain sensors as warn.

Requirements

• haive init must have been run in the workspace (creates .ai/memories/)
• Node.js ≥ 20 (for the haive CLI, optional — extension reads files directly)

Installation

From VSIX (local build)

cd packages/vscode
pnpm install
pnpm build
npx vsce package
code --install-extension haive-vscode-*.vsix

From Marketplace

Search hAIve in the Extensions view, or run:

ext install hAIve.haive-vscode

Then initialize hAIve in your repo (see Getting started).

Settings

Keybindings

Commands

All commands are available from the Command Palette (Ctrl/Cmd+Shift+P).

Browse & search

Author & curate

Observability & quality

Setup

🌱 Curating seeds: stack-pack seeds are generic starter knowledge kept at background priority. The sidebar groups unanchored seeds under Seeds — needs curation; anchor one to a real file (or replace it with a repo-specific note) to make it high-signal for agents.