Instruction file imported from JavaStudentAlex/european-defense-tech-hackathon-berlin (.github/instructions/agent_maintenance_workflow.instructions.md). Copyright stays with the author.

Agent Maintenance Workflow

Purpose

Use this when editing AGENTS.md, .github/instructions, .github/agents, .github/skills, or repository automation that prompts coding agents (.github/workflows/jules_next_task.yml, .github/workflows/jules_auto_merge.yml, .github/scripts/resolve_gsd_slice.py).

Maintenance Rules

• Keep instruction files task-scoped and repository-specific to the multi-sensor track fusion challenge: feed normalization, uncertainty modeling, geometric association, clutter rejection, weighted fusion, conservative classification, confidence scoring, operator priority, and the FastAPI/GeoJSON demo.
• Do not add arbitrary tool dependencies that are not in pyproject.toml.
• Keep Python version references aligned with .python-version and pyproject.toml.
• Project-scoped skills belong under .github/skills.
• Remove copied instructions from unrelated project domains unless the active project plan explicitly reintroduces them.
• Commands written for cloud agents (Jules) and CI must run without local-only wrappers such as rtk; the rtk prefix applies only to local sessions where the wrapper is installed.

Review Checklist

• Agents correctly address normalization, association, fusion, clutter/priority, and API/GeoJSON responsibilities for fuse(feeds) in src/fusion.py.
• Documentation commands align with Python 3.13, uv, Ruff, pytest, and the pre-commit manual-stage quality batch.
• Fixed contracts stay protected: agents must not modify src/sources.py (input) or src/output.py (output) without an explicit user request.
• No files refer to copied repository domains unrelated to multi-sensor track fusion.