ja-zoe/tokeneyezed

How ja-zoe/tokeneyezed works

Tokeneyezed is a long-horizon coding-agent harness: it repeatedly plans and runs attempts toward a scored CommonMark-rendering goal, retaining useful history across restarts and agent handoffs. Its main workflow is orchestrated by the controller; Atlas-backed memory informs bounded briefs and records attempt results, while the observer gates tool actions and filters events. Separate product branches provide split-based scoring and retrieval evaluation, plus a read-only dashboard for comparing and inspecting runs. Wiring details are drawn from the README and sampled modules; unsampled implementation links are omitted where ownership is unclear.

Main entry points: graph.py Run graph · service.py Observer service · retrieval_eval.py Retrieval evaluation · cli.py Run CLI

Components 23

Run controller

Run CLI src/tokeneyezed/controller/cli.py
Run graph src/tokeneyezed/controller/graph.py
Attempt planner src/tokeneyezed/controller/planner.py
Agent runner src/tokeneyezed/controller/runners

Atlas memory

Attempt brief src/tokeneyezed/data/brief.py
Goal state src/tokeneyezed/data/goals.py
Attempt ledger src/tokeneyezed/data/ledger.py
Data writes src/tokeneyezed/data/writes.py

Observer

Hook shim src/tokeneyezed/observer/shim.py
Observer service src/tokeneyezed/observer/service.py
Pre-action gate src/tokeneyezed/observer/core.py
Post-action review src/tokeneyezed/observer/postcheck.py
Mongo event writer src/tokeneyezed/observer/storage.py

Evaluation

Split scorer src/tokeneyezed/eval/scorer.py
Retrieval evaluation src/tokeneyezed/data/retrieval_eval.py

Dashboard

Dashboard UI public/app.js
Dashboard API src/tokeneyezed/dashboard/server.py
Dashboard reads src/tokeneyezed/data/dashboard_reads.py

Other components

Operator
Coding agent
CommonMark spec
MongoDB Atlas
Voyage AI

Connections 29

Operator → Run CLI: starts or resumes
Run CLI → Run graph: runs session
Run graph → Goal state: seeds goals
Run graph → Attempt brief: builds brief
Attempt brief → MongoDB Atlas: reads memory
Run graph → Attempt planner: plans attempt
Attempt planner → MongoDB Atlas: requests model
Run graph → Agent runner: runs attempt
Agent runner → Coding agent: launches
Coding agent → CommonMark spec: implements
Run graph → Split scorer: scores attempt
Split scorer → CommonMark spec: validates against
Run graph → Attempt ledger: updates ledger
Attempt ledger → MongoDB Atlas: persists attempts
Run graph → MongoDB Atlas: checkpoints run
Coding agent → Hook shim: emits hook events
Hook shim → Observer service: posts events
Observer service → Pre-action gate: checks pre-actions
Observer service → Post-action review: checks post-actions
Observer service → Mongo event writer: persists verdict
Mongo event writer → Data writes: inserts event
Data writes → MongoDB Atlas: writes event
Operator → Retrieval evaluation: runs evaluation
Retrieval evaluation → MongoDB Atlas: queries attempts
Retrieval evaluation → Voyage AI: embeds queries
Operator → Dashboard UI: views runs
Dashboard UI → Dashboard API: requests API
Dashboard API → Dashboard reads: routes reads
Dashboard reads → MongoDB Atlas: reads run data

More Python diagrams: tarun7r/vocal-agent · temadol/gmail_generator · 5758703/cv_pythonvue_tigerpro · andreaskuhnle/simrlfab · cloudbase/coriolis

Drawn by GitDiagram on September 26, 2026 from the repository's file tree, README and source files. Read as Markdown ·