* Rebase notification preference fix onto upstream main * fix(runtime): publish a failed artifact delivery on the run stream (#8252) * fix(runtime): publish the delivery failure on the run stream * docs(api): note the delivery-failure stream event * test(runtime): cover the deferred delivery-failure frame * docs(runtime): log the delivery fix and explain the fallback exemption * test(e2e): present the recorded output in the replay scenario The write/read-file fixture predates the delivery check, so the replayed run has been ending as an error without presenting note.txt. This PR makes that failure visible on the stream, which broke the replay golden and the render e2e. Add the present_files step the current scenario produces and regenerate the golden. * docs: drop the CHANGELOG entry CONTRIBUTING.md asks contributors to leave CHANGELOG.md and CHANGELOG_zh.md to the maintainers at merge time, so the Unreleased ordering and the English/Chinese pair stay in sync. --------- Co-authored-by: 一片一片 <3037678834@qq.com>
4.1 KiB
Backend (backend/)
The Python backend: the publishable harness in packages/harness/deerflow/, the
public extension contract in packages/extension-api/, and the unpublished
Gateway and IM channels in app/. In every deployment the agent runtime runs
inside the Gateway. Design docs are in docs/; start with
ARCHITECTURE.md and API.md.
Guides by area
Read the guide for an area before you change it.
| Area | Guide |
|---|---|
| Run streaming, checkpoints, run ownership | runtime |
| Repositories, leases, scheduled tasks, schema changes | persistence, migrations |
| Scheduler failure isolation | design conventions |
config.yaml, extensions_config.json |
config |
| Lead agent, prompt, thread state, middleware chain | agents, middlewares |
| Memory, subagents, skills, tools | memory, subagents, skills, tools |
| Sandbox, uploads, MCP, models, extensions | sandbox, uploads, mcp, models, extensions |
| Trace ID, embedded client, other root modules | harness package |
| Search, fetch and sandbox providers | community, then the provider's own guide |
| Gateway, routers, IM channels | gateway, routers, channels |
| Tests, benchmarks | tests, benchmarks |
Verify
make test # offline suite in parallel shards
make test-blocking-io # strict blocking-I/O suite
make test-bundled-extensions # behavior suites of bundled examples (not in make test)
make lint
make format
make test-live # calls real APIs; needs DEER_FLOW_RUN_LIVE_TESTS=1
Run make test and make test-blocking-io before and after a change. Every
feature and bug fix ships with a test in tests/. Live tests stay opt-in, so the
default suite needs no network and no API key.
Rules
-
The harness never imports
app.app.*importsdeerflow.*, and not the reverse, because the harness is published on its own.tests/test_harness_boundary.pyenforces this in CI. -
Start the Gateway with
make devormake gateway.make devexcludesDEER_FLOW_HOMEandbackend/sandboxfrom the reload watcher. A bareuvicorn --reloadrestarts the Gateway during a run, because agent tasks write Python files belowDEER_FLOW_HOME. -
From the repository root, run backend tools with
uv run --no-sync --project backend.--projectkeeps the caller's working directory, so relative config paths still resolve.--no-synckeeps installed extras such as the PostgreSQL drivers. -
Run Python through
uv runinsidebackend/, without--no-sync, so the environment matches the lock. -
Guard POSIX-only test markers with
os.name. Windows collects the same files. -
Tool approvals need addressable calls. Strip missing, empty and duplicate tool-call IDs before creating a pause or a tool result. See the run interaction policy.
-
Add a schema change with
make migrate-rev MSG="..."and follow the migrations guide. -
Notification preferences are a send-time trust boundary. Malformed provider or per-event resolver output must fail closed and remain retryable without consuming an attempt; never treat malformed data as opt-in.