▚▚ fix(hooks): resolve the output dir from GRAPHIFY_OUT in the sh gates, not a literal graphify-out/

[breken-ai]

With GRAPHIFY_OUT set to anything but the default (the worktree / shared-output setup from #686), graphify hook install produces a post-checkout hook that never rebuilds: switching branches prints nothing and writes nothing to ~/.cache/graphify-rebuild.log. The post-commit hook has the mirror problem — a commit that only touches custom-out/graph.json (tracked outputs) launches a full rebuild, which is exactly the loop the "only graph artifacts changed" guard was added to prevent.

Root cause

#1423 moved the Python rebuild bodies onto os.environ.get('GRAPHIFY_OUT', 'graphify-out'), but the shell that runs before them still hardcodes the directory name:

• _CHECKOUT_SCRIPT: if [ ! -d "graphify-out" ]; then exit 0; fi — with a renamed output dir this is always true, so the hook exits before _PYTHON_DETECT and the launcher ever run.
• _HOOK_SCRIPT: _NON_GRAPH=$(echo "$CHANGED" | grep -v '^graphify-out/' || true) — paths under the configured dir are not filtered, so a graph-only commit looks like a source change.

The existing test only asserts that the Python body mentions GRAPHIFY_OUT, so the sh gates in front of it were never pinned.

Fix

• graphify/hooks.py (_CHECKOUT_SCRIPT): gate on ${GRAPHIFY_OUT:-graphify-out} (trailing slash stripped), the same source the rebuild body reads.
• graphify/hooks.py (_HOOK_SCRIPT): build _NON_GRAPH with a POSIX case loop keyed on "$_GFY_OUT"/* instead of grep -v, so the dir name is matched literally rather than as a regex. Each pattern carries the optional leading ( because bash 3.2 (macOS /bin/sh) otherwise mis-parses the pattern's ) as the end of the $(...) substitution — that was caught by the test run, not guessed.

Tests

tests/test_hooks.py — five tests that run the emitted hook scripts under a real sh in a scratch git repo, with the pinned python replaced by a stub that passes the find_spec probe and swallows the launcher, so the tests observe whether the hook reaches the launch line rather than running a rebuild:

• test_checkout_hook_rebuilds_when_graphify_out_is_renamed — fails on HEAD (no launch line).
• test_commit_hook_skips_graph_only_commit_when_graphify_out_is_renamed — fails on HEAD (launch line present).
• Controls that pass before and after: checkout still no-ops when neither dir exists; the default graphify-out/ graph-only commit is still skipped; a commit with a source file alongside the renamed dir still launches.

Reverting only hooks.py makes the first two fail again.

Not changed

• _PYTHON_DETECT's second probe still reads graphify-out/.graphify_python literally. It is one of five fallbacks for locating an interpreter (the pinned python is tried first), and no writer of that file lives in graphify/, so it is left alone rather than widened without a witness.
• An absolute GRAPHIFY_OUT that points inside the repo (/repo/custom-out) still is not relativised for the commit filter; git diff --name-only emits repo-relative paths, so that case behaves as it did before this change (a rebuild is launched). Relative names and the default are the documented forms and are what this fixes.
• fix(hooks): resolve merge-driver output path from .graphifyrc, skip gitignored graph.json #3207 (persisting GRAPHIFY_OUT to .graphifyrc so it survives a fresh shell) is complementary: once the value is available in the hook's environment, these gates now honour it.

---

▚▚ Shipped by breken — self-healing software. This one's on us. breken.ai

[graphify-labs]

Graphify reviewed this change.

Worth a look — the grounded gate found no coupling regressions or blocking issues, but 2 advisory finding(s) below merit a look before merge.

---

Graphify review — findings

Makes the shell gates in front of the rebuild honour GRAPHIFY_OUT instead of a hardcoded graphify-out/: post-commit's "only graph artifacts changed" filter now strips paths under the configured output dir (via a POSIX case loop that survives bash 3.2's $(...) parsing), and post-checkout's existence check resolves the same dir before deciding whether a graph was ever built. Without this, a renamed output dir made the branch-switch rebuild a silent no-op and let a graph-only commit trigger a full rebuild. Adds sh-level tests that run the emitted hooks with the launcher stubbed to assert the rename and default-name paths both launch (or skip) correctly.

Worth a look

• Post-commit graph-only filter misses ./-prefixed GRAPHIFY_OUT — graphify/hooks.py:424 · Escalate · medium
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review
• Post-commit GRAPHIFY_OUT filter does not normalize equivalent relative paths — graphify/hooks.py:429 · Escalate · medium
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 416 functions depend on the 227 functions this change touches.

Health — this change adds coupling hotspots:

• new: install() — 37 callers, 7 callees
• new: dispatch_command() — 2 callers, 124 callees
• new: dispatch_install_cli() — 2 callers, 31 callees
• new: status() — 8 callers, 6 callees
• new: uninstall() — 9 callers, 5 callees
• new: uninstall_all() — 2 callers, 13 callees
• new: test_poisoned_manifest_is_healed() — 0 callers, 6 callees

Verification — 416 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 242 function(s) in the blast radius were not formally verified this run

Test selection

Test selection

21 of 275 test file(s) selected (8%) via static blast radius.

• tests/test_affected_cli.py — impact
• tests/test_agents_platform.py — impact
• tests/test_codebuddy.py — impact
• tests/test_devin.py — impact
• tests/test_explain_cli.py — impact
• tests/test_extract_cli.py — impact
• tests/test_global_add_tag_inference.py — impact
• tests/test_god_nodes_cli.py — impact
• tests/test_hollow_chunks_arm_shrink_guard.py — impact
• tests/test_hook_chain_survives_skip.py — impact
• tests/test_hooks.py — impact, changed-test
• tests/test_incomplete_build_guard.py — impact
• tests/test_install.py — impact
• tests/test_install_references.py — impact
• tests/test_merge_chunks_validation.py — impact
• tests/test_multigraph_diagnostics.py — impact
• tests/test_no_dedup_flag.py — impact
• tests/test_path_cli.py — impact
• tests/test_query_cli.py — impact
• tests/test_query_induced_edges.py — impact
• tests/test_unverified_semantic_shrink.py — impact

Selection is safe under the controlled-regression assumption; always-run tests + a periodic full run are the backstops. Advisory — it never changes the check verdict.

· 7 more finding(s) on lines outside this diff (see the check run).

[safishamsi]

Shipped in v0.9.62 (now on PyPI: uv tool install graphifyy==0.9.62). Your commits were cherry-picked with authorship preserved, so this shows up under your GitHub contributions. Thanks @breken-ai!