equilateral-agents

EquilateralAgents Open Core

22 self-learning AI agents. Build institutional knowledge that compounds over time. MIT licensed.

Transform your AI coding assistant into a learning system that gets smarter with every mistake you make (and prevents you from making it again).

---

What's New in v3.1.0

YAML Standards Format (v3.0.0 - Breaking Change)

All standards now use YAML format instead of markdown. This enables machine-readable standards that agents consume directly:

id: lambda-database-standards
category: serverless
priority: 10
rules:
  - action: ALWAYS
    rule: "Cache single database client at module scope for warm start reuse"
  - action: NEVER
    rule: "Use connection pools in Lambda - Lambda handles one request at a time"
anti_patterns:
  - "Creating new Pool() per invocation"
tags: [lambda, database, cost-optimization]

StandardsLoader (v3.0.0)

New core utility that loads YAML standards from a three-layer directory hierarchy:

const { StandardsLoader } = require('equilateral-agents-open-core');
const loader = new StandardsLoader({ projectRoot: process.cwd() });

const all = await loader.loadAll();           // All standards from all layers
const security = await loader.loadByTags(['security']); // Filter by tag
const rules = await loader.getRulesForAgent('SecurityReviewerAgent'); // Agent-specific

Three layers (later overrides earlier):

1. .standards/yaml/ - Official open standards (submodule)
2. .standards-community/ - Community-contributed patterns (submodule)
3. .standards-local/ - Your team's conventions (git-ignored)

Bundled Project/Object Skill (v3.1.0)

Session memory and standards injection included by default. Closes the loop between YAML standards on disk and enforcement in every AI session:

1. On session start: Reads ~/.project-object/{project}/context.md and injects prior session context
2. Standards loading: Scans .standards/yaml/*.yaml, extracts rules, injects as [REQUIRE]/[AVOID]/[PREFER] directives
3. On session end: Harvests new decisions, patterns, and corrections from the transcript
4. Cross-platform sync: Context syncs to Claude Code, Cursor, Codex, Windsurf via project-object sync

Migration from v2.x

• Rename .standards-local/*.md files to .yaml and convert to YAML schema
• StandardsContributor now generates .yaml output
• Knowledge harvest reports output .yaml instead of .md
• See CHANGELOG.md for complete migration details

---

Why EquilateralAgents?

The Problem: Codebases Don't Learn

Traditional development:

• ❌ Same security bugs discovered 3+ times
• ❌ N+1 query performance issues in every new feature
• ❌ Production incidents from patterns you've seen before
• ❌ New developers repeat mistakes the team already solved
• ❌ No institutional memory - knowledge lives in people's heads

The Solution: A Learning System

EquilateralAgents creates a feedback loop:

1. Execute Workflows (agents scan your code)
        ↓
2. Agent Memory (tracks what worked, what failed)
        ↓
3. Knowledge Harvest (extract patterns weekly)
        ↓
4. Create Standards (document "What Happened, The Cost, The Rule")
        ↓
5. Enforce Standards (AI checks before changes, agents validate)
        ↓
6. Fewer Incidents (prevent repeating mistakes)
        ↓
[Loop back to step 1]

Result: Your codebase gets smarter over time. Mistakes happen once, not repeatedly.

---

Perfect For

🌱 Greenfield Projects

Start right from day 1:

• Security scanning before first commit
• Quality gates before bad patterns take root
• Document decisions as you make them
• Build standards library alongside code

Example journey:

• Week 1: Run security/quality workflows, create first standards
• Month 1: 10+ standards covering your specific domain
• Month 3: New feature? Check standards first. AI references them automatically.

🏗️ Brownfield Codebases

Fix systematically, not randomly:

• Agents identify patterns across entire codebase
• Document each fix as a standard (prevent recurrence)
• Gradually eliminate entire classes of bugs
• Track progress: incidents per month going down

Example journey:

• Week 1: Security scan finds 50 issues. Fix 10, document pattern.
• Month 2: Similar issue caught by agent during PR. Standard working.
• Month 6: That entire class of bugs eliminated from codebase.