alex314618-create/jevrev

How alex314618-create/jevrev works

JevRev is a decision layer alongside a host LLM agent, exposed through a CLI/JSON protocol and a human-facing TUI. The main workflow explores candidate approaches (Sift), evaluates probe evidence and chooses a direction, then audits successive rounds of work (Loop); Long independently monitors extended sessions for risks. The README documents these product stages, while sampled source confirms selected CLI, evidence, audit, persistence, and TUI interactions. Some subsystem wiring is therefore shown only where supported by those excerpts or documentation.

Main entry points: cli.ts CLI and JSON · store.ts Event journal · campaign.ts Probe campaign · commands.ts Loop commands

Components 26

Host interface

CLI and JSON src/cli.ts

JevSift

Shadow activation src/activation.ts
Judge questions src/questions.ts
Jev judge src/judge.ts
Sift policy src/policy.ts
Probe campaign src/workflow/campaign.ts
Probe evidence src/workflow/evidence.ts
Candidate decision src/workflow/decide.ts

JevLoop

Command recorder src/evidence/recorder.ts
Evidence bundles src/evidence/store.ts
Loop commands src/loop/commands.ts
Round auditor src/loop/auditor.ts
Next action src/loop/planner.ts
Loop state src/loop/store.ts

JevLong

Long commands src/long/commands.ts
Event normalization src/long/normalize.ts
Event journal src/long/store.ts
Risk signals src/long/signals.ts
Alert policy src/long/policy.ts
Session watch src/long/watch.ts
Loop bridge src/long/bridge.ts

Session view

Terminal cockpit src/tui/app.ts
Session discovery src/tui/sessions.ts
View preferences src/tui/store.ts

Other components

Host agent
Jev service

Connections 30

Host agent → CLI and JSON: submits requests
CLI and JSON → Shadow activation: evaluates request
CLI and JSON → Probe campaign: builds campaign
Probe campaign → Sift policy: filters candidates
Probe campaign → Probe evidence: prepares evaluations
Probe evidence → Candidate decision: supplies evidence
Candidate decision → Jev judge: requests judgment
Jev judge → Jev service: scores questions
Judge questions → Jev judge: provides questions
Host agent → Command recorder: records probe results
Command recorder → Evidence bundles: writes observations
CLI and JSON → Loop commands: dispatches commands
Loop commands → Round auditor: audits round
Round auditor → Evidence bundles: reads evidence
Loop commands → Loop state: loads and updates
Round auditor → Next action: assesses outcome
Next action → Loop state: persists next action
CLI and JSON → Long commands: dispatches commands
Long commands → Event normalization: normalizes events
Long commands → Event journal: appends and reads
Session watch → Event journal: reads session events
Session watch → Risk signals: checks for risks
Risk signals → Alert policy: provides signals
Alert policy → Event journal: stores alert state
Loop bridge → Event journal: imports Loop events
CLI and JSON → Terminal cockpit: opens cockpit
Terminal cockpit → View preferences: reads and writes settings
Terminal cockpit → Session discovery: refreshes sessions
Session discovery → Loop state: reads Loop state
Session discovery → Event journal: reads Long state

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·