Skip to content

ejentum/agent-teams

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Teams

A library of multi-agent teams (agentic workflows) using Ejentum's RA²R Logic API as the cognitive layer. Each team is a deployable artifact: drop it into your runtime, configure one credential, run.

The teams in this repo are tuned for specific tasks where multi-cognitive analysis genuinely beats single-agent output. They use cross-lab agent diversity (Anthropic, Google, Alibaba, Zhipu, OpenAI) to reduce correlated failure modes that come from any single model family.


Catalog

A 4-agent team (architect + reasoner + implementer + reviewer) for heym v0.0.9+ that performs deep adversarial code review. Refuses to rubber-stamp PRs, sources every concern from a specialist agent, grounds approvals in specific positive evidence. Each specialist applies a distinct Ejentum cognitive harness (reasoning, code, anti-deception). → Setup and verification tests

Evaluation framework for comparing baseline LLM output against Ejentum-augmented LLM output, with a third-party blind judge. Available as n8n workflows, a Python CLI, and integration patterns for agentic IDEs (Cursor, Antigravity, Claude Code). The instrument is the artifact: import, run on your own prompts, see the diff. → Spec and integrations


How these teams use Ejentum

Each agent (or evaluation step) calls the Ejentum Logic API in one of four cognitive modes:

Mode Purpose
reasoning General reasoning scaffold — failure-mode suppressors, target patterns, falsification tests
code Code-specific failure suppressors — cross-layer mismatch, environment drift, invariant violations
anti-deception Suppresses sycophancy, hallucination, prompt injection, false certainty
memory Perception sharpening, behavioral calibration, cross-turn observation

Two paths to wire each cognitive mode into a specialist agent:

  • HTTP Request tool (canonical): the curl pattern shown in each team's README. Works on every runtime.
  • MCP server (when the runtime supports MCP clients): install ejentum-mcp via Smithery and the four harnesses appear as harness_* tools your specialist can call directly. No HTTP wiring per agent.

Get an Ejentum API key at ejentum.com/pricing. Free tier: 100 calls total.

Read more at ejentum.com/docs.


Adding a new team

The pattern (orchestrator + N specialists, each with a mode-specific harness) generalizes. To contribute or fork a new team:

  1. Pick a target task with multi-cognitive value (refactor planning, security audit triage, production debug forensic, spec verification, test design from traces, etc.).
  2. Decide which Ejentum modes each specialist needs.
  3. Build the workflow on your runtime of choice (heym, n8n, LangGraph, your own).
  4. Drop into a new folder here with: README.md, workflow file(s), skill MDs (if applicable), system prompts, screenshots.
  5. Add a row to the Catalog above.

Future teams in the planning phase: refactor planner, security audit triage, production debug forensic.


License

MIT. See LICENSE.

Contact

info@ejentum.com

About

A library of multi-agent teams using Ejentum cognitive harnesses. Drop-in deployable agent teams for code review, evaluation, and more.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors