Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
45 lines (36 loc) · 1.7 KB

CONTRIBUTING.md

File metadata and controls

45 lines (36 loc) · 1.7 KB

Contributing

Thanks for looking. agentlift is small and deliberately layered — the front half is pure, the back half is the only place that touches the network.

Layout

src/agentlift/
  model.py            data classes (Project, AgentSpec, SkillSpec, McpServerSpec)
  parser.py           folder -> Project           (pure)
  planner.py          Project -> DeployPlan        (pure)
  lockfile.py         idempotency state
  anthropic_target.py DeployPlan -> API calls      (network)
  runtime.py          run a deployed agent / the same folder locally
  cost.py             token -> USD estimate
  graders.py          substring + LLM graders
  cli.py              argparse entry point

Dev setup

python -m pip install -e ".[dev]"
pytest -m "not live"          # fast, no key
ANTHROPIC_API_KEY=... pytest -m live   # hits the real API, costs a few cents

Ground rules

  • Keep parser.py and planner.py pure. No network, no clock, no randomness. If a behavior can be tested offline, it belongs there and gets an offline test.
  • Every translation rule needs an offline test asserting the plan (tests/test_planner.py). The plan is the contract.
  • New API behavior gets confirmed live first, then encoded. Don't guess wire format from docs alone — the beta moves.
  • Surface, don't swallow. Anything agentlift can't translate becomes a Diagnostic (error or warning), visible in agentlift plan.

Adding a provider target

Implement a new target with the same apply(plan) contract as anthropic_target.Deployer. The planner already emits provider-agnostic operations; a target maps them to that provider's API. Keep the convention identical so one folder deploys anywhere.