Reygent Workflow

Visual diagrams of the reygent workflow, decision flows, and agent interactions.

Full Workflow Overview

The complete reygent workflow from spec input to reviewed pull request:

Planner Clarification Loop

How the planner resolves ambiguous specs:

Implementation Stage

Dev and QE agent execution based on mode:

Test Gate Retry Flow

What happens when tests fail:

Security Review Decision Flow

PR Creation Flow

How reygent creates a pull request:

Branch Naming Convention

Branch names use conventional commit prefixes. Type is auto-detected from issue metadata or prompted interactively.

Valid branch types: feat, fix, chore, refactor, docs, test, style, perf

Agent Spawning Internals

How each agent subprocess works:

Resuming a Failed Run

After every completed stage, reygent run writes a snapshot to .reygent/runs/<runId>.json capturing the run options and the accumulated TaskContext. If a run fails or is interrupted, reygent continue replays from that snapshot.

Key behaviors:

• Snapshots are listed as "unfinished" if any stage failed or the final pr-review stage hasn't completed.
• The picker shows the 5 most recent unfinished runs, sorted by last update time.
• CLI flags on continue (--auto-approve, --verbose) override the snapshot's stored values; everything else (provider, model, type, security threshold, retry count) is inherited from the original run.
• Snapshots are not auto-pruned after success — clean up .reygent/runs/ manually if it grows.

Complete Data Flow

How TaskContext flows through the reygent workflow: