Repository navigation
feat(install): add graphify vibe install for Mistral Vibe CLI - #2537
xavierpestel-ai wants to merge 4 commits into
Conversation
e138e96 to
a02e474
Compare
![graphify-labs[bot]](https://avatars.githubusercontent.com/in/4356270?s=60&v=4){kind=small}
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This pull request appears to reintroduce version 0.9.33 as "unreleased" in the CHANGELOG, removing the previously listed 0.9.34 and 0.9.35 entries, and adds a new feature line describing Mistral Vibe support. It updates the README with graphify vibe install documentation across the install tables and command reference, and wires up new Vibe-related imports (install/uninstall hooks, hook path helpers) in __main__.py. The build.py change removes the _coerce_hyperedge_member_refs helper. The changed-symbols list spans a much broader surface than the shown diff—covering skillgen expected fixtures/fragments, extractor engine/resolution logic, CLI and install modules, and multiple test files (serve, watch, cli_export)—suggesting the full changeset also touches skill generation output, extraction/resolution code, and their associated tests.
No blocking issues surfaced.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 5383 functions depend on the 3268 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 33 callees - worse:
install()— 5 callers, 12 callees - worse:
_build_server()— 2 callers, 21 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - …and 1 more
Verification — 5383 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: 5383 function(s) in the blast radius were not formally verified this run
· 2 grounded finding(s) anchored inline below; 7 more finding(s) on lines outside this diff (see the check run).
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR adds Mistral Vibe (mistralai/mistral-vibe) as a supported platform for graphify vibe install. It introduces a new vibe platform config, a skill file with user-invocable frontmatter, AGENTS.md integration, and two pre_tool hooks (for grep and read_file) merged into vibe's .vibe/hooks.toml via tomlkit, with corresponding install/uninstall helper functions exported from __main__.py. The surface area includes new install helpers in graphify/install.py (_vibe_install, _vibe_uninstall, _install_vibe_hook, _uninstall_vibe_hook, _vibe_hooks_path, _vibe_hook_entries, plus supporting path/normalization utilities), updates to README and CHANGELOG documentation, and changes to skill-generation tooling/fragments and vibe-related tests. Hook ownership is tracked by name, and both global (VIBE_HOME) and project scopes are handled.
No blocking issues surfaced. 6 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 880 functions depend on the 605 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 33 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 880 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: 880 function(s) in the blast radius were not formally verified this run
· 2 grounded finding(s) anchored inline below; 6 more finding(s) on lines outside this diff (see the check run).
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR adds Mistral Vibe (mistralai/mistral-vibe) as a supported platform for graphify install. It introduces a new vibe platform config, a skill file with user-invocable frontmatter, an AGENTS.md section, and install/uninstall logic that merges two pre_tool hooks (for grep and read_file) into vibe's hooks.toml using tomlkit, honoring VIBE_HOME for global scope and both project/global paths. The surface area spans install.py (new _vibe_* helpers, hook entries, config entry, dispatch in install()), exports in __main__.py, README/CHANGELOG documentation updates, associated skill-generation fragments, and tests covering the vibe dispatch and uninstall behavior.
No blocking issues surfaced. 8 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 880 functions depend on the 605 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 33 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 880 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: 880 function(s) in the blast radius were not formally verified this run
· 2 grounded finding(s) anchored inline below; 6 more finding(s) on lines outside this diff (see the check run).
a02e474 to
f8758ff
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.
Graphify review — findings
This PR adds Mistral Vibe as a supported platform for graphify install. It introduces a new vibe platform config, a skill-vibe.md skill file, install/uninstall functions (_vibe_install, _vibe_uninstall), and pre_tool hook management (_install_vibe_hook, _uninstall_vibe_hook) that merge two hooks (for grep and read_file) into vibe's hooks.toml using tomlkit. It also wires up VIBE_HOME-aware path resolution, updates the CLI dispatch and __main__ exports, and adds README/CHANGELOG documentation plus corresponding tests. The surface area spans install.py (new functions and platform config), __main__.py (imports), documentation files, skill/fragment generation tooling, and a test module covering hook installation, comment preservation, and user-authored hook ownership.
Worth a look
- install(platform='vibe') ignores strict flag —
graphify/install.py:612· 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 — 886 functions depend on the 611 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 33 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 886 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: 886 function(s) in the blast radius were not formally verified this run
· 2 grounded finding(s) anchored inline below; 6 more finding(s) on lines outside this diff (see the check run).
f8758ff to
caa1460
Compare
|
Fixed in The bot was right — Two changes:
Added regression test Verified end-to-end:
|
There was a problem hiding this comment.
Graphify reviewed this change.
Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.
Graphify review — findings
This PR adds Mistral Vibe (mistralai/mistral-vibe) as a supported platform for the graphify vibe install / uninstall commands. It introduces a vibe entry in the platform config plus new helper functions for resolving the vibe home directory (honoring VIBE_HOME), destination paths for skill/AGENTS.md files, and merging two pre_tool hooks (grep + read) into .vibe/hooks.toml via tomlkit, with name-based ownership and a --strict option. It also wires these into the installer dispatch, exports the new symbols from __main__.py, and updates the README and CHANGELOG, alongside generated skill fragments and tests.
Worth a look
- _refuse_to_modify does not stop execution, causing UnboundLocalError on unparseable hooks.toml —
graphify/install.py· Escalate · high- 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 — 888 functions depend on the 613 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 888 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: 888 function(s) in the blast radius were not formally verified this run
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
caa1460 to
a3ab759
Compare
|
Investigated in The Process exits 1, file preserved, no That said, the finding surfaced two real cosmetic issues worth fixing:
Both changes are in the single line pair at Verified: 29 vibe tests pass, 558 install-suite tests pass, 5/5 skillgen CI checks pass, actual malformed-TOML repro exits 1. Regarding the partial-install ordering (skill + AGENTS.md land before |
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR adds Mistral Vibe as a new supported install target for the graphify vibe install (and --platform vibe) commands. It introduces a vibe platform config plus helper functions to resolve vibe's home dir (honoring VIBE_HOME), write the skill file/AGENTS.md, and merge two name-owned pre_tool hooks (matching grep and read_file) into .vibe/hooks.toml using tomlkit. It also wires the new symbols into __main__.py, updates the README and CHANGELOG, extends the skill-generation fragments/expected outputs, and generalizes the "refuse to modify" path to cover TOML alongside JSON, with accompanying tests. Surface area touched: graphify/install.py, graphify/__main__.py, the skillgen tooling and its expected fixtures, the vibe skill/graphify skill files, tests (test_vibe), README, and CHANGELOG.
No blocking issues surfaced. 5 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 888 functions depend on the 613 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 888 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: 888 function(s) in the blast radius were not formally verified this run
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
a3ab759 to
e225076
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR adds Mistral Vibe as a supported install target for graphify. It introduces a new graphify vibe install / uninstall flow (plus a --platform vibe alias) that writes a SKILL.md skill file, an AGENTS.md section, and two pre_tool hooks (matching grep and read_file) into vibe's hooks.toml, with support for global vs. project scope and a --strict option. The changes span the installer dispatch in graphify/install.py (new _vibe_* helpers for path resolution, hook entry construction via shlex/tomlkit, and merge/uninstall logic), new exports in __main__.py, CHANGELOG and README documentation entries, and updates to the skill generator (tools/skillgen) fragments and expected artifacts. The _refuse_to_modify helper was also adjusted to mention TOML and to raise SystemExit instead of sys.exit. Surface area includes the CLI dispatch, install/uninstall code paths, skillgen rendering/artifacts, and associated tests (tests/test_vibe.py).
No blocking issues surfaced. 6 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 905 functions depend on the 624 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 905 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: 905 function(s) in the blast radius were not formally verified this run
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
e225076 to
5c25f01
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR adds Mistral Vibe as a supported platform for the graphify install/vibe install commands. It introduces a new vibe entry in the platform config, a dedicated skill file, vibe-specific path resolution (honoring VIBE_HOME), and installer/uninstaller functions that merge two pre_tool hooks (for grep and read_file) into vibe's hooks.toml using tomlkit, plus an always-on AGENTS.md section. It also wires up the new exports in __main__.py, updates the README/CHANGELOG documentation, and adjusts the shared _refuse_to_modify error handling to mention TOML and raise SystemExit.
No blocking issues surfaced. 8 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 917 functions depend on the 626 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 917 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: 917 function(s) in the blast radius were not formally verified this run
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
5c25f01 to
1df5ed7
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Formal verification. 1 change(s) alter behavior, breaking input(s) attached.
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
Graphify review — findings
This PR adds Mistral Vibe (mistralai/mistral-vibe) as a supported platform for graphify install/uninstall. It introduces a dedicated vibe install path that writes a SKILL.md skill file, an AGENTS.md section, and two pre_tool hooks (matching grep and read_file) into vibe's hooks.toml, with support for both global (VIBE_HOME) and project-scoped installs plus a --strict flag. The surface area includes new helper functions in graphify/install.py (hook entry building, TOML merging via tomlkit, path resolution, install/uninstall), corresponding exports in __main__.py, README/CHANGELOG documentation, and new tests plus skillgen fragments/expected outputs for the vibe skill content.
No blocking issues surfaced. 5 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 918 functions depend on the 627 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 918 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: 918 function(s) in the blast radius were not formally verified this run
Formal verification
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
Could not verify: Could not verify \_platform\_skill\_destination.
The verifier did not have enough to check \_platform\_skill\_destination, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 9 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly KeyError — names the real obstacle, not a sampling gap)
Could not verify: Could not verify \_project\_install.
The verifier did not have enough to check \_project\_install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_project\_uninstall.
The verifier did not have enough to check \_project\_uninstall, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_refuse\_to\_modify.
The verifier did not have enough to check \_refuse\_to\_modify, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `settings_path` is annotated `Path` — outside the synthesizable primitive/collection set
No difference found (not proven): No behavior difference found in dispatch\_install\_cli (not a proof).
The verifier ran both versions of dispatch\_install\_cli on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
Could not verify: Could not verify install.
The verifier did not have enough to check install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 9 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly SystemExit — names the real obstacle, not a sampling gap)
No difference found (not proven): No behavior difference found in monolith\_roundtrip (not a proof).
The verifier ran both versions of monolith\_roundtrip on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
1df5ed7 to
585ac6c
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Formal verification. 1 change(s) alter behavior, breaking input(s) attached.
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
Graphify review — findings
This PR adds Mistral Vibe (mistralai/mistral-vibe) as a supported platform for the graphify install/vibe install flow. It introduces vibe-specific paths and config handling — a SKILL.md skill file, an AGENTS.md section, and two pre_tool hooks (matching grep and read_file) merged into .vibe/hooks.toml via tomlkit — along with global-scope resolution honoring VIBE_HOME, a --strict option, and name-based hook ownership for idempotent install/uninstall. Surface area includes new install/uninstall functions in graphify/install.py, corresponding exports in graphify/__main__.py, README and CHANGELOG documentation, expected skill/test fixtures, and skillgen tooling updates. There's also a small change switching a hook-install abort from sys.exit(1) to raise SystemExit(1) and updating the related error message to mention TOML.
No blocking issues surfaced. 6 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 919 functions depend on the 628 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 919 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: 919 function(s) in the blast radius were not formally verified this run
Formal verification
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
Could not verify: Could not verify \_platform\_skill\_destination.
The verifier did not have enough to check \_platform\_skill\_destination, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 9 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly KeyError — names the real obstacle, not a sampling gap)
Could not verify: Could not verify \_project\_install.
The verifier did not have enough to check \_project\_install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_project\_uninstall.
The verifier did not have enough to check \_project\_uninstall, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_refuse\_to\_modify.
The verifier did not have enough to check \_refuse\_to\_modify, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `settings_path` is annotated `Path` — outside the synthesizable primitive/collection set
No difference found (not proven): No behavior difference found in dispatch\_install\_cli (not a proof).
The verifier ran both versions of dispatch\_install\_cli on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
Could not verify: Could not verify install.
The verifier did not have enough to check install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 9 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly SystemExit — names the real obstacle, not a sampling gap)
No difference found (not proven): No behavior difference found in monolith\_roundtrip (not a proof).
The verifier ran both versions of monolith\_roundtrip on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
585ac6c to
5ee7f43
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.
Graphify review — findings
This PR adds Mistral Vibe ("vibe") as a supported install target for the graphify vibe install / graphify vibe uninstall commands, alongside the existing platforms. The changes register vibe in the platform config and skill-destination logic (global scope honoring VIBE_HOME, project scope under .vibe/), and add functions to install/uninstall two pre_tool hooks (grep and read_file matchers) into a hooks.toml file using tomlkit, with name-based ownership and a --strict variant. Documentation (README, CHANGELOG) and __main__.py exports are updated to reflect the new commands and symbols. Surface area touched: CHANGELOG.md, README.md, graphify/__main__.py, graphify/install.py, plus skillgen tooling/fragments/expected outputs and vibe-related tests.
Worth a look
- _install_vibe_hook error message references non-existent package name 'graphifyy' —
graphify/install.py:1567· 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 — 920 functions depend on the 629 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
_project_uninstall()— 5 callers, 14 callees - worse:
dispatch_install_cli()— 2 callers, 34 callees - worse:
install()— 5 callers, 12 callees - new:
_vibe_install()— 4 callers, 9 callees - worse:
_copy_skill_file()— 9 callers, 4 callees - worse:
_project_install()— 2 callers, 14 callees - worse:
uninstall_all()— 2 callers, 14 callees - new:
_vibe_uninstall()— 4 callers, 5 callees
Verification — 920 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: 920 function(s) in the blast radius were not formally verified this run
· 3 grounded finding(s) anchored inline below; 5 more finding(s) on lines outside this diff (see the check run).
|
+1 for this feature from a (potential) new user. I was considering using graphify, noticed it didn't support mistral, and considered opening an issue to ask whether you'd be open to a PR implementing support. My pre-search turned up this PR. So, thanks to @xavierpestel-ai for putting it together - I'll probably try your fork later this week until this is merged ... |
|
I've noticed a few issues while testing this from the fork. @xavierpestel-ai - is it better to list them here or as issues on your fork? |
Could you add them here ? |
|
Sub-agent selection for update / new graph - recommendation in skill ideally needs updating to reflect standard vibe agents: The explore agent that vibe defaults to based upon the skill instructions does not have file-write or edit permissions: and at another point |
|
project-level |
5ee7f43 to
a2f6092
Compare
There was a problem hiding this comment.
Graphify reviewed this change.
Worth a look — the grounded gate found no coupling regressions or blocking issues, but 5 advisory finding(s) below merit a look before merge.
Formal verification. 1 change(s) alter behavior, breaking input(s) attached.
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
Graphify review — findings
Adds full-parity Mistral Vibe support to graphify vibe install / install --platform vibe, writing a user-invocable skill file (exposes /graphify as a native slash command), an always-on AGENTS.md section, and two name-owned pre_tool hooks in .vibe/hooks.toml (matchers grep and read_file) that nudge toward graphify query; global scope honors VIBE_HOME, hooks.toml merges preserve hand-authored comments via tomlkit, and --strict blocks the first raw read per session. Registers a writable graphify-extract subagent at <scope>/agents/graphify-extract.toml via _install_vibe_extract_agent, scoping write_file to **/graphify-out/.graphify_chunk_*.json so semantic extraction can actually emit chunk JSON — the built-in read-only explore subagent silently produced zero chunks and aborted the pipeline. uninstall_all and graphify vibe uninstall sweep the hooks, agent, skill, and AGENTS.md section symmetrically.
Worth a look
- Non-project install path passes unsupported strict keyword —
graphify/install.py:2518· Escalate · high- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review
- Unquoted/unescaped INPUT_PATH interpolated into shell and Python source enables injection —
graphify/skill-vibe.md· Escalate · high- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review
- All-cached semantic path skips required semantic merge file —
graphify/skill-vibe.md· Escalate · high- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review
- User path is embedded in Python source without escaping —
graphify/skill-vibe.md:119· Escalate · high- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review
- Vibe hook install follows hooks.toml symlink outside project —
graphify/install.py:1667· 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 — 1003 functions depend on the 677 functions this change touches.
Health — this change adds coupling hotspots:
- new:
dispatch_command()— 2 callers, 124 callees - new:
codebuddy_install()— 20 callers, 5 callees - new:
claude_install()— 19 callers, 4 callees - new:
_project_uninstall()— 5 callers, 14 callees - new:
gemini_install()— 10 callers, 7 callees - new:
dispatch_install_cli()— 2 callers, 34 callees - new:
claude_uninstall()— 17 callers, 4 callees - new:
_copy_skill_file()— 13 callers, 5 callees - …and 20 more — each is listed as a finding
Verification — 1003 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: 1003 function(s) in the blast radius were not formally verified this run
Formal verification
Behavior changes: uninstall\_all changes behavior, here is the input that shows it.
The verifier found a concrete input on which uninstall\_all behaves differently before and after the change. If that change is intended, ship it; if not, this is your bug.
Guarantee: This difference was REPRODUCED, the verifier actually ran both versions on that input and saw them disagree. It is real, not an artifact.
Evidence: On input \{"project\_dir":"0","purge":"True"\}, the old code produced None but the new code produces raises RuntimeError. Paste that input straight into a regression test.
No difference found (not proven): No behavior difference found in dispatch\_install\_cli (not a proof).
The verifier ran both versions of dispatch\_install\_cli on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
Could not verify: Could not verify install.
The verifier did not have enough to check install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 23 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly SystemExit — names the real obstacle, not a sampling gap)
Could not verify: Could not verify \_platform\_skill\_destination.
The verifier did not have enough to check \_platform\_skill\_destination, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: not verifiable: all 23 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly KeyError — names the real obstacle, not a sampling gap)
Could not verify: Could not verify \_project\_install.
The verifier did not have enough to check \_project\_install, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_project\_uninstall.
The verifier did not have enough to check \_project\_uninstall, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `project_dir` is annotated `Path | None` — outside the synthesizable primitive/collection set
Could not verify: Could not verify \_refuse\_to\_modify.
The verifier did not have enough to check \_refuse\_to\_modify, so it is saying so rather than guessing. No false assurance is the whole point.
Guarantee: No guarantee either way, this is an honest abstention, not a pass.
Note: Reason: parameter `settings_path` is annotated `Path` — outside the synthesizable primitive/collection set
No difference found (not proven): No behavior difference found in monolith\_roundtrip (not a proof).
The verifier ran both versions of monolith\_roundtrip on many inputs and saw identical behavior every time. Strong evidence the change is safe, but evidence, not a proof.
Guarantee: Empirical: differential testing (both versions run on many generated inputs). A divergence on an untested input remains possible, so this is 'no counterexample found', not 'proven equivalent'.
Note: An input the sampler did not try could still differ.
· 4 grounded finding(s) anchored inline below; 24 more finding(s) on lines outside this diff (see the check run).
|
Hi everybody, is there anything I can help with to bring this forward? would be very appreciated to have support in vibe. Or those who work on it, and plans to fix the review points? Thanks! |
a2f6092 to
7725e9b
Compare
|
Picking this back up. The branch was 444 commits behind Changes in this push1. Rebase onto current
2.
3. Sub-agent selection (already fixed in
Verification
@bzed — thanks for the nudge, the branch should be current now. @MusicalNinjaDad — the two issues you hit from the fork should be resolved; if you get a chance to re-test, the key things to verify are (1) |
Vibe's project-scope install dir (.vibe/) holds graphify's own skill, agent, and hook artifacts — scanning it wastes semantic extraction on config files instead of real code. Add .vibe to _SKIP_DIRS so it is pruned by default, matching how .graphify and graphify-out are handled.
|
I'll try to take a look at it next week. Thanks |
Adds Mistral Vibe (github.com/mistralai/mistral-vibe) as a full-parity platform integration, matching Claude Code's shape: skill file + AGENTS.md always-on section + hooks.toml pre_tool guards. Vibe reads .vibe/skills (project, trusted) and VIBE_HOME/skills (user global, default ~/.vibe/skills) per its Agent Skills-compliant harness manager. `user-invocable: true` in the skill frontmatter exposes /graphify as a native slash command in vibe's autocomplete. Two pre_tool hooks (grep + read_file) nudge toward `graphify query` instead of raw filesystem tools, mirroring the claude/gemini/codex guard pattern. The --strict flag flows to the read_file guard only, matching claude --strict. Matchers use vibe's actual snake-cased tool names (Grep -> grep, ReadFile -> read_file per vibe/core/tools/base.py:get_name) since vibe's hook matcher uses fnmatch. Ownership is decided by hook `name` in a fixed set, NOT a substring match on `command` -- a substring match would silently delete a user's own hooks that shelled out to graphify for unrelated reasons. Uses tomlkit for hooks.toml merges so hand-authored comments, key ordering, and unrelated [[hooks]] entries survive graphify install/upgrade. Command strings are built with shlex.join so exe paths containing shell metacharacters are quoted safely for vibe's asyncio.create_subprocess_shell executor. Honors the VIBE_HOME env var for all three artifacts (skill, hooks, AGENTS.md) so users on non-default vibe setups don't get orphaned installs into a directory vibe never reads. Skillgen registration: vibe is a monolith bucket (its SkillMetadata schema requires extra frontmatter fields user-invocable + allowed-tools that _render_frontmatter emits only name+description for). Post-v8 monoliths opt out of monolith_roundtrip via roundtrip_ref=None (matches _v8_baseline_ref's existing post-v8 handling for the agents platform). CLI: graphify vibe install # global VIBE_HOME (default ~/.vibe/) graphify vibe install --project # project ./.vibe/ + ./AGENTS.md graphify vibe install --strict # block first raw read per session graphify vibe uninstall # symmetric cleanup graphify install --platform vibe # alias for the global install graphify uninstall # uninstall_all sweeps vibe too Tests: 28 covering CLI paths, VIBE_HOME symmetry, name-based ownership, shlex-safe quoting across 9 hostile paths, 4 malformed TOML shapes, and idempotency. Verified end-to-end against real Mistral Vibe 2.24.0: skill discovery, /graphify slash-command execution, pre_tool hooks fire on grep and read_file invocations, and full pipeline runs on a sample repo producing a queryable graph. Adds tomlkit>=0.13 as a runtime dep (comment-preserving TOML round-trip; stdlib tomllib is read-only and needs 3.11+ while requires-python is 3.10).
Vibe's built-in explore subagent is read-only (enabled_tools omits write_file), so Task-dispatched semantic extraction inheriting it silently produced zero .graphify_chunk_NN.json files and aborted the pipeline. Register a graphify-extract subagent alongside the skill that enables write_file scoped to **/graphify-out/.graphify_chunk_*.json only (no unbounded write grant), reuses the explore system prompt, and is symmetrically installed/uninstalled by _install_vibe_extract_agent / _uninstall_vibe_extract_agent. Skill fragment updated so Task calls dispatch with subagent_type="graphify-extract" on vibe. uninstall_all sweeps the agent alongside the rest.
…latform enum Two small post-rebase fixes: 1. _vibe_install called _refresh_all_version_stamps, which Graphify-Labs#2694 (rebased onto v8) removed in favor of per-platform stamping. _copy_skill_file already writes the stamp for the vibe skill, so replace the call with a comment noting the v8 behavior. 2. install() error message enumerated 'gemini, cursor' as the special cases but not 'vibe', giving misleading help text when a user passes an unknown platform. Add 'vibe' to the enumeration.
Vibe's project-scope install dir (.vibe/) holds graphify's own skill, agent, and hook artifacts — scanning it wastes semantic extraction on config files instead of real code. Add .vibe to _SKIP_DIRS so it is pruned by default, matching how .graphify and graphify-out are handled.
7725e9b to
a05a4f5
Compare
This branch has not been deployed
3 participants
Summary
Adds Mistral Vibe (
mistralai/mistral-vibe) as a full-parity platform integration, matching Claude Code's shape: skill file +AGENTS.mdalways-on section +hooks.tomlpre_toolguards.What's included
CLI
Artifacts installed
VIBE_HOME/skills/graphify/SKILL.md(or./.vibe/skills/…in project scope) withuser-invocable: truefrontmatter — exposes/graphifyas a native slash command in vibe's autocomplete.AGENTS.mdsection with the shared## graphifymarker (uses graphify's existing_replace_or_append_sectionhelper, matches how other AGENTS.md platforms register).pre_toolhooks inhooks.toml— matchersgrepandread_file(vibe's real snake-cased tool names pervibe/core/tools/base.py:get_name) that nudge towardgraphify queryfor search / raw-read operations.Design decisions worth flagging
Hook matchers use vibe's snake-cased class names. Vibe's hook matcher uses
fnmatch, and vibe'sBaseTool.get_name()snake-cases the class (Grep→grep,ReadFile→read_file). Matching on"read"alone would silently never fire against the realread_filetool — this was caught during real-vibe integration testing before landing.VIBE_HOMEhonored for all three artifacts. Vibe's harness manager readsVIBE_HOME/skills,VIBE_HOME/hooks.toml, andVIBE_HOME/AGENTS.md; hardcoding~/.vibewould silently orphan installs for users who setVIBE_HOME=/opt/vibe-shared.Ownership by hook
name, not command substring. Filtering existing[[hooks]]entries by"graphify" in commandwould silently delete a user's own hook that happens to shell out tographifyfor unrelated reasons. Filtering by hookname in {"graphify-nudge-search", "graphify-nudge-read"}is precise.shlex.joinfor hook commands. Vibe runs hooks viaasyncio.create_subprocess_shell, so metacharacters in the exe path ($, backticks,;,(,)) parse as shell operators. The previous "quote only if space" heuristic (still used by the older claude/gemini hook installers) leaves injection open on macOS home dirs containing parentheses.tomlkitforhooks.tomlmerges. Preserves user comments and unrelated[[hooks]]entries across install/upgrade cycles. Stdlibtomllibis read-only and only 3.11+ (graphify supports 3.10), sotomlkit>=0.13is added as a runtime dep — small pure-Python wheel.Skillgen registration as a monolith. Vibe's
SkillMetadataschema requires extra frontmatter fields (user-invocable,allowed-tools) that skillgen's_render_frontmatteremits onlyname+descriptionfor. Registering as a monolith bucket (likeaideranddevin) lets the skill file carry its own frontmatter without extending the shared renderer. Post-v8 monoliths (no pristine baseline to freeze against) opt out ofmonolith_roundtripviaroundtrip_ref = None— mirrors_v8_baseline_ref's existing post-v8 handling for theagentsplatform (4-line skillgen change).uninstall_allsweep.graphify uninstallnow removes vibe artifacts alongside every other platform — otherwise vibe hooks would keep firing after users think they've cleaned up.Files changed (12)
graphify/install.py_PLATFORM_CONFIG["vibe"],_platform_skill_destinationbranch,_install_vibe_hook/_uninstall_vibe_hook,_vibe_install/_vibe_uninstallorchestrators,_vibe_home,_vibe_hooks_path,_vibe_agents_md_path,_vibe_hook_entries,_normalize_toml, CLI dispatch branch,_CLI_INSTALL_COMMANDSentry,_project_install/_project_uninstallbranches,uninstall_alladditiongraphify/__main__.py_vibe_install,_vibe_uninstall,_install_vibe_hook,_uninstall_vibe_hook,_vibe_hooks_path,_vibe_hook_entries)graphify/skill-vibe.mdtests/test_vibe.pytools/skillgen/fragments/core/vibe.mdtools/skillgen/expected/graphify__skill-vibe.mdtools/skillgen/platforms.tomlroundtrip_ref)tools/skillgen/gen.pymonolith_roundtripopt-out for post-v8 monolithspyproject.tomltomlkit>=0.13runtime dep,skill-vibe.mdin package-data, description enumerationREADME.mdCHANGELOG.md0.9.33 (unreleased)uv.lockuv syncTesting
Automated
tests/test_vibe.pycovering CLI paths,VIBE_HOMEsymmetry, name-based ownership,shlex.join-safe quoting across 9 hostile paths, 4 malformed TOML shapes, idempotency (no.graphify-bakchurn on re-install), strict-flag flow.test_ollama.py,test_ollama_retry_cap.py,test_labeling.py::test_label_communities_batches_when_over_batch_size).--check,--audit-coverage,--schema-singleton,--monolith-roundtrip,--always-on-roundtrip).End-to-end against real Mistral Vibe 2.24.0
vibe -p "/graphify --help"executes the skill body verbatimgrepandread_filetool invocations (verified via a disk-tracer hook command that touches a file when triggered)graphify path "handle_login()" "execute()"traced the full login flow (handle_login → login → find_user_by_credentials → execute) across 4 modulesChecklist
v8(active development branch per README)feat: <description>per README §Git workflowuv run pytest tests/ -qpassestools/skillgen/platforms.tomland the blessedexpected/fixture updated0.9.33 (unreleased)descriptionenumeration includes Mistral Vibe