AI Agents

AI coding agents work from the same documentation and architecture decisions as human developers. AI agents are not the source of truth — Git repositories and project documentation are.

Tools

Tool	Status
Claude Code	Operational installed on the VPS
OpenAI Codex CLI	Operational installed and authenticated on the VPS
ChatGPT ↔ GitLab Self-Managed	Deferred unavailable in the current ChatGPT workspace

Agent instructions

• PROJECT_CONTEXT.md (root of hoggar-docs) — global context, read first.
• CLAUDE.md — present in hoggar-docs, hoggar-infra and hoggar-web.
• AGENTS.md — optional.

Core rules

Before modifying Hoggar, an agent must:

1. Read PROJECT_CONTEXT.md completely, then the repository's CLAUDE.md and/or AGENTS.md.
2. Read relevant architecture documentation and ADRs.
3. Inspect the repository before proposing changes.
4. Not assume a planned technology is already installed.
5. Not introduce infrastructure without justification.
6. Never expose, print, store or commit secrets.
7. Prefer small, reviewable changes and run relevant checks.
8. Propose an ADR when changing an established architectural decision.

When asked to analyze the project, agents do not modify anything unless explicitly instructed.

Future roles

Potential agent roles later: architect, backend-spring, frontend-angular, security-reviewer, test-reviewer, docs-writer, docs-translator. They are not created yet.