agent-borg 3.2.0

Collective memory for AI coding agents — your agent learns from every session

Borg

Your AI coding agent forgets everything between sessions. Borg remembers.

---

10-Second Demo

Paste an error. Get a structured fix.

$ borg debug "ModuleNotFoundError: No module named 'cv2'"

============================================================
ERROR: ModuleNotFoundError: No module named 'cv2'
============================================================
[dependency-resolution] (Python)
Problem: Missing system-level dependency masquerading as pip issue

ROOT CAUSE:
  Category: environment-mismatch
  opencv-python requires system libs that pip can't install alone

INVESTIGATION TRAIL:
  1. [first] requirements.txt
     → Check if opencv-python or opencv-python-headless is listed
     grep: opencv
  2. [then] Dockerfile or system packages
     → Confirm libgl1-mesa-glx is installed for GUI builds

RESOLUTION SEQUENCE:
  1. Install headless variant (no system deps needed)
     Command: pip install opencv-python-headless
     Why: Avoids libGL dependency entirely
  2. If GUI needed, install system deps first
     Command: apt-get install -y libgl1-mesa-glx

ANTI-PATTERNS (don't do these):
  ✗ pip install opencv-python without system deps
    Fails because: ImportError at runtime even though pip succeeds

EVIDENCE: 47/52 successes (90%) over 127 uses
         Avg resolve time: 2.3 min
============================================================

That's not a template. That's learned from real agent sessions.

---

30-Second Setup

pip install agent-borg
borg debug "your error message here"

That's it. No API keys. No config. No account.

---

Why

Every AI coding agent — Claude Code, Cursor, Cline, Windsurf — starts from scratch every session. It doesn't know what worked last time. It doesn't know what failed.

Borg is collective memory. When one agent solves a problem, every agent learns. When one agent fails, nobody repeats the mistake.

• Agent hits an error → borg debug returns the fix
• Agent starts a task → borg observe returns how to approach it
• Agent needs patterns → borg search finds what worked before
• Export to your platform → borg generate writes the rules file

---

Features

• Instant debugging — paste any error, get root cause + fix + anti-patterns
• Works offline — no API calls, no cloud, runs locally
• Collective learning — fixes improve from real agent outcomes
• Platform export — one command to generate .cursorrules, .clinerules, CLAUDE.md, or .windsurfrules
• 17 MCP tools — plug into any MCP-compatible agent
• Task guidance — get step-by-step approaches before you start coding
• Pattern search — find what worked across all sessions
• Failure memory — tracks what didn't work so agents stop repeating mistakes

---

Platform Setup

Claude Code

borg generate systematic-debugging --format claude
# Creates CLAUDE.md in your project

Cursor

borg generate systematic-debugging --format cursor
# Creates .cursorrules in your project

Cline

borg generate systematic-debugging --format cline
# Creates .clinerules in your project

Windsurf

borg generate systematic-debugging --format windsurf
# Creates .windsurfrules in your project

MCP (any compatible agent)

{ "mcpServers": { "borg": { "command": "borg-mcp" } } }

---

Quick Start

# 1. Debug an error
borg debug "TypeError: Cannot read properties of undefined (reading 'map')"

# 2. Get task guidance before you start
borg observe "refactor authentication to use JWT tokens"

# 3. Search for patterns that worked
borg search "docker networking"

# 4. Export rules to your editor
borg generate systematic-debugging --format cursor

# 5. Classify an error without full guidance
borg debug --classify "ECONNREFUSED 127.0.0.1:5432"

---

How It Works

Borg ships with packs — structured knowledge extracted from real debugging sessions. Each pack contains:

• Problem signature — what the error looks like
• Root cause — why it actually happens
• Investigation trail — where to look, in order
• Resolution sequence — exact commands to fix it
• Anti-patterns — what not to do (and why it fails)
• Evidence — success rate from real usage

Packs improve over time. When agents report outcomes via borg feedback, success rates update and better approaches surface.

---

Install

pip install agent-borg          # core
pip install agent-borg[crypto]  # with signing support
pip install agent-borg[all]     # everything

Requires Python 3.10+.

---

License

MIT