TinySDLC – Agent orchestrator with SDLC role discipline for AI coding (MIT)

1 point

3 months ago

Problem: when multiple AI coding agents collaborate, there's no structure — no separation of duties, no handoff discipline, no traceability.

TinySDLC is a minimal agent orchestrator that adds SDLC role discipline to AI coding. 8 roles (researcher, architect, coder, reviewer, tester, etc.), each with isolated workspaces, scoped tool permissions, and enforced separation of duties. Coder can't self-approve. Reviewer can't be bypassed.

Design constraints: file-based queue (zero external dependencies), multi-channel (Discord, Telegram, WhatsApp, Zalo), security hardening (7 credential scrubbing patterns, 12 injection patterns blocked, 8 shell deny patterns).

Protocol-first: roles are contracts, AI providers are swappable.

Companion repo MTS-SDLC-Lite contains the governance methodology (Spec → Gate → Evidence → Approval) — templates, playbooks, processes. Both MIT.

Result of 12 iterations across 5 production projects.

GitHub: https://github.com/Minh-Tam-Solution/tinysdlc

Methodology docs: https://github.com/Minh-Tam-Solution/MTS-SDLC-Lite

Both MIT licensed. Feedback and criticism welcome.