Documentation
¶
Overview ¶
Package e2e holds end-to-end tests that drive the real no-slop binary against a temporary git repo and a fake agent. Tests live behind the `e2e` build tag so they are excluded from `go test ./...` and only run via `make e2e` (or `go test -tags=e2e ./internal/e2e/...`).
The fake agent (cmd/fakeagent) is symlinked under each agent's binary name (claude, codex, opencode) into a temp PATH directory, and replies with deterministic canned responses defined by Scenario YAML or by the built-in "everything is clean" default. Every invocation is appended to $FAKEAGENT_LOG so tests can assert on which prompts the pipeline made.
Why e2e at all: see the audit in the PR description. Unit tests for pipeline orchestration, executor branching, and CLI wiring give weak signal compared to actually running the binary against a real git repo. The e2e suite consolidates the "user pushes a branch and the pipeline runs to completion" journey into a single high-coverage test that is meant to grow rather than fan out into many small e2e files.
Temporary daemon ownership: each NewHarness acquires a slot in internal/e2edaemon (inventory + concurrency cap). scripts/e2e.sh and TestMain reap inventoried daemons when Cleanup cannot run. External sleep-loop keepalives are out of scope. See internal/e2edaemon/doc.go for the SIGKILL recovery boundary.
Source Files