streb streb

Architecture

A Complete Framework for Agentic Software Delivery

streb is the operating framework for a structured, agentic software delivery process — where requirements, tasks, code, and audit trails are interconnected, automated, and governed by default.

Mode-Agnostic Execution

Interactive and Headless — same environment

Both execution modes share the same container image, the same CLAUDE.md rules, the same hooks, and the same toolchain. Only the entry point and output format differ.

Interactive Mode
Developer-facing

Claude Code launches in your terminal or IDE. You work conversationally with the agent, issue follow-up prompts, and use /ralph-loop for multi-iteration autonomous tasks. You decide when the session ends.

$ streb run interactive
Claude Code connected to container environment
All tools, Beads, requirements graph available
> Implement the session timeout feature (req-7a2b)
Headless Mode
CI/pipeline-facing — no human present

The agent is invoked with a prompt. The Ralph Wiggum Loop iterates until COMPLETE or BLOCKED. A structured result.json is written to .agent-output/. CI reads it to determine pass or fail.

$ streb run "Implement req-7a2b" \
--max-iterations 10
COMPLETE — result.json written
# status, bead-id, artifacts, iterations

CI Integration

Label an issue. The agent implements it.

streb includes a GitHub Actions workflow template with three trigger paths. When a GitHub issue receives the agent label, the workflow automatically starts a headless agent run using the issue content as the prompt.

The agent implements the task, writes tests, commits with the task ID, and writes a structured result. The workflow passes or fails based on the agent's COMPLETE or BLOCKED signal.

Manual dispatch with a custom prompt
Automatic trigger on agent label
Optional nightly cron for maintenance tasks
.github/workflows/agent.yml
# Three trigger paths
on:
workflow_dispatch:
inputs: { prompt: ... }
issues:
types: [labeled]
schedule:
# - cron: '0 2 * * *'
# Run agent
- run: streb run
"$PROMPT"
--max-iterations 15
# Read result
- run: cat .agent-output/result.json
status: COMPLETE
bead_id: bd-a1b2c3
iterations: 4

Roadmap

From setup tool to delivery operating system

v0.8 · Available Now
  • Complete dev environment setup in <5 minutes
  • Claude Code + Beads + Dagger + templates
  • Interactive and Headless execution modes
  • Ralph Wiggum Loop for autonomous iteration
  • Jira bidirectional sync
  • GitHub Actions workflow templates
Coming Next
  • Deep task tracking: agent state ↔ Beads integration
  • Requirements traceability (streb req commands)
  • Queryable audit trail (streb audit)
  • Production-grade CI execution engine
  • Shared config management (streb config pull/push)
  • Drift detection for stale requirements

Ready to set up your environment?

One command. Under 5 minutes. Start building with a governed, auditable AI dev environment today.

Developer Setup Guide

Deploying to a team?

See how to standardize streb across your practice — with VDI support, compliance guidance, and Jira integration.

Team Deployment Guide