Skip to content

AGENTS.md

Latest commit

 

History

History
32 lines (23 loc) · 1.61 KB

AGENTS.md

File metadata and controls

32 lines (23 loc) · 1.61 KB
Raw

Development Notes

This repo is a focused long-running worker, not a broad assistant distribution.

Active Surface

  • Runtime package: nipux_cli/
  • Tests: tests/nipux_cli/
  • Entry point: nipux
  • State home: ~/.nipux or NIPUX_HOME
  • Planning notes: plans/nipux-runtime-notes.md

Constraints

  • Keep the default tool surface small and explicit.
  • Do not reintroduce broad upstream surfaces such as gateways, skills, plugins, web UI, ACP, RL environments, voice, image generation, or arbitrary terminal execution. The chat-first terminal UI is part of Nipux's active product surface; keep it generic, minimal, and backed by persisted worker state.
  • Preserve restartability: every worker step should persist state before and after tool execution.
  • Store exact evidence as artifacts. Summaries should point back to artifacts instead of replacing them.
  • Keep memory_index entries compact and artifact-referenced; do not use raw transcript replay as the long-term state strategy.
  • Prefer OpenAI-compatible model serving, configured through ~/.nipux/config.yaml.
  • Keep runtime behavior domain-neutral. Do not add task-specific or environment-specific guards, keyword lists, examples, prompts, tools, or tests and describe them as generic framework improvements.

Validation

Use the focused suite:

PYTEST_ADDOPTS='' uv run --extra dev python -m pytest -q
uv run --extra dev ruff check --isolated nipux_cli tests/nipux_cli

Use nipux daemon --once --fake for a no-model smoke test. Use nipux logs JOB_ID --verbose or nipux watch JOB_ID --verbose when inspecting what a background job is actually doing.