argus-standards 0.1.0

AI Agent Engineering Standards Layer

argus-standards

Argus injects engineering discipline into AI coding agents via platform-specific instruction files generated from composable standards packs.

AI coding agents skip TDD, produce commit blobs, and violate SOLID by default. Argus fixes this by generating the right system prompt files for your platform — one command, any combination of standards.

Install

pip install argus-standards

Quick Start

1. Create .argus.yml in your project root:

packs:
  - tdd
  - atomic-commit
  - solid

2. Generate platform files:

argus generate

Argus writes instruction files for every supported platform it detects in your project. For example, with Claude Code and Cursor present:

✓ AGENTS.md
✓ CLAUDE.md
✓ .claude/rules/tdd.md
✓ .claude/rules/atomic-commit.md
✓ .claude/rules/solid.md
✓ .claude/skills/tdd/SKILL.md
✓ .claude/skills/atomic-commit/SKILL.md
✓ .claude/skills/solid/SKILL.md
✓ .cursor/rules/tdd.md
✓ .cursor/rules/atomic-commit.md
✓ .cursor/rules/solid.md

Commands

Command	Description
argus init	Create a starter .argus.yml in the current directory
argus generate	Generate platform files from .argus.yml
argus generate --dry-run	Preview files that would be written without writing them
argus generate --check	Exit non-zero if generated output differs from disk (for CI)
argus packs list	List all available packs
argus packs show <name>	Show the full content of a pack
argus platforms list	List all supported platforms
argus validate	Validate .argus.yml without generating files

Packs

Packs are composable engineering standards. Include only the ones your project needs.

Pack	Description
tdd	Test-driven development — RED → GREEN → COMMIT → REFACTOR cycle
atomic-commit	One logical change per commit, conventional commit format
solid	SOLID principles with mechanically detectable violation checklist
code-quality	Cyclomatic complexity ≤ 10, method ≤ 20 lines, class ≤ 300 lines
pre-commit	Non-negotiable pre-commit gate combining all active standards
error-handling	Exception hierarchy, raise vs return rules, catching at boundaries
type-safety	Full annotation coverage, mypy-clean, no Any in signatures
documentation-standards	Docstrings on public API, imperative mood, no what-comments
dependency-injection	Constructor injection, depend on abstractions, single composition root
design-patterns	Strategy/Factory/Observer patterns to eliminate type-dispatch
refactoring	Safe refactoring cycles, code smell catalog, preparatory refactoring
testing-strategy	Test pyramid, test doubles guide, right layer for each test type

Platforms

Platform	Files Generated
Claude Code (claude)	CLAUDE.md, .claude/rules/{pack}.md, .claude/skills/{pack}/SKILL.md
Cursor (cursor)	.cursor/rules/{pack}.md
GitHub Copilot (copilot)	.github/copilot-instructions.md
OpenCode (opencode)	opencode.json, .opencode/skills/{pack}/SKILL.md, .opencode/commands/{pack}.md

All platforms also receive AGENTS.md (OpenAI Codex / general agents).

Development Setup

git clone https://github.com/pmurasky/argus.git
cd argus
pip install -e ".[dev]"
pytest

License

MIT © Peter Murasky Jr