Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
57 lines (42 loc) · 1.76 KB

CONTRIBUTING.md

File metadata and controls

57 lines (42 loc) · 1.76 KB
Raw

Contributing

AgentGuard Graph has a narrow contract: normalize local evidence, build an agent security graph, analyze attack paths, and report unknowns. Contributions must preserve the local-first, read-only, no-telemetry boundary.

Setup

pip install -e ".[dev]"
python -m compileall -q src tests
python -m unittest discover -s tests -v
coverage run --source=src/agentguard_graph -m unittest discover -s tests
coverage report -m --fail-under=85

The scanner runtime should stay standard-library-only unless the project explicitly accepts a new dependency.

Rules

  • Do not add network calls to the core scan path.
  • Do not execute commands, code, prompts, tool descriptors, or evidence fields.
  • Treat all evidence strings as untrusted.
  • Escape HTML output.
  • Keep reports self-contained.
  • Report missing information as visibility gaps.
  • Keep scoring explainable with named dimensions, controls, and caps.
  • Add tests for parser, rule, report, and CLI changes.

Evidence Adapters

Adapters import offline exports or static local config.

Expected behavior:

  • accept JSON first
  • preserve source metadata
  • assign confidence explicitly
  • keep inferred mappings conservative
  • emit warnings for incomplete exports
  • avoid secrets in samples and docs
  • include fixture tests and an E2E path when collect behavior changes

Findings And Scoring

New rules need:

  • stable rule id
  • graph node and edge references
  • evidence summary
  • unknowns and visibility gaps
  • recommended next evidence
  • remediation
  • score dimensions and controls
  • tests for positive and controlled cases

Documentation

Update README.md, docs/E2E_WORKFLOW.md, docs/ROADMAP.md, and CHANGELOG.md when changing user-facing commands, evidence formats, report fields, or supported workflows.