agentic-swe-grounding-system

Modular, self-referencing Markdown grounding system for agentic AI software engineering, architecture, planning, refactoring, implementation, review, verification, and operations.

This repository is optimized for AI agents, not linear human reading.

Agent Use

Agents should not load the full repository by default.

Default loading sequence:

1. Load 00-kernel.md.
2. Load 01-routing-index.md.
3. Select one primary workflow SOP.
4. Load only triggered supporting modules.
5. Use templates only when producing artifacts.
6. Use checklists before completion.
7. Report verification evidence.

Repository Shape

• 00-kernel.md: invariant operating law.
• 01-routing-index.md: task-to-module router.
• 02-loading-policy.md: context budget rules.
• workflows/: operational SOPs.
• architecture/: agent-friendly design patterns.
• agent-harness/: tool/action/observation/memory harness guidance.
• prompt-engineering/: prompt contracts and task brief structure.
• testing-evals/: tests, evals, and verification gates.
• security/: safety, privacy, prompt/tool injection, and approval gates.
• templates/: copy-paste artifacts.
• checklists/: concise completion gates.
• anti-patterns/: failure diagnosis and recovery.

Prime Rule

Every token must route, constrain, verify, template, or recover.