testing-gibson-harnesslisted

# Testing the Gibson harness (shell-only, no UI) There is no web UI and no test suite: everything is bash + plain node. Test with the shell tool directly; do not record a screen video. ## Scratch target repo (2 minutes) ```bash rm -rf /tmp/scratch && mkdir /tmp/scratch && cd /tmp/scratch git init -q -b main && git config user.email t@t.co && git config user.name tester echo 'def add(a,b): return a-b' > calc.py printf 'import calc\nassert calc.add(2,2)==4\n' > test_calc.py git add -A && git commit -qm init git checkout -qb feat/x && echo 'def add(a,b): return a-b # wip' > calc.py && git commit -qam wip ``` A repo with **no git remote** is useful on purpose: it exercises the `SLUG` basename fallback in `devin-supervisor.sh`. Scripts create `<repo>/gibson/` themselves. ## Vendor CLI auth is the usual blocker — check it FIRST ```bash codex exec --sandbox read-only --cd /tmp/scratch 'say ok' # 401 if unauthenticated claude -p --output-format text --permission-mode plan 'say ok' # "Not logged in" grok -p 'say ok' # "Not signed in" / 403 no credits ``` Auth for these is typically only on the user's own Mac mini. An `XAI_API_KEY` may be present yet still fail with `403 permission-denied: … team doesn't have any credits` — an API key existing does not mean tokens are available. Report a blocked tier as blocked-by-auth, and prove the harness plumbing separately (below) rather than claiming a live-model pass. ## Reading the captured r