This directory holds the executable spec for the debug-reach skill. If this test passes, the skill's documented happy path works against a real nvim+tmux+debugpy stack.

test_breakpoint_flow.py drives the full guyide debug surface against a freshly-installed NvGuy + vscodium.nvim + nvim-dap + debugpy:

1. guyide doctor -> ready
2. guyide debug list-configs -> sees sample-debugpy
3. guyide debug break set --file sample.py --line 13
4. guyide debug start --config sample-debugpy
5. guyide debug state --wait --reason breakpoint --vars --frames
6. assert a and b are reported as locals at the stop
7. guyide debug continue -> hit again on the next loop iteration
8. guyide debug stop -> session_active becomes false

Every step is asserted to carry the "schema":"guyide/v1" envelope.

python -m pip install pytest pytest-timeout debugpy
pytest -xvs testdata/e2e/

Optional env vars:

• GUYIDE_E2E_NVGUY_LOCAL=/path/to/nvguy — use a local clone instead of fetching from GitHub.
• GUYIDE_E2E_NVGUY_REF=<branch> — pin a different NvGuy ref (default: main).
• GUYIDE_E2E_KEEP=1 — keep the ephemeral tmux server alive after the test for inspection.

The following nvim-launch.debug-rpc Lua issues will cause the test to fail until they are patched in vscodium.nvim:

1. list_configs() calls launch_json.load_configs which does not exist (likely renamed). Step 2 will fail.
2. step_over/into/out blindly return {success: true} even when no dap session exists. Not exercised by this test, but worth tracking.

These are tracked in BRANCH-NOTES.md of the feature/cli-skill-bridge branch.

.github/workflows/e2e.yml runs this on every push, every PR, and nightly at 03:17 UTC. Diagnostics from a failing run are uploaded as the e2e-diagnostics artifact.