Repository navigation
chore(ci): add argot voice check + config - #39
Open
damienmeur wants to merge 1 commit into
Open
damienmeur wants to merge 1 commit into
damienmeur wants to merge 1 commit into
Conversation
Non-blocking argot voice check on PRs into main: sticky score comment and SARIF code-scanning annotations. Commit the fitted voice config (argot.toml) and ignore personal overrides (argot.local.toml).
damienmeur
added a commit
to get-tmonier/argot
that referenced
this pull request
Jul 13, 2026
argot fit scaffolded a default argot.toml (untracked) and appended argot.local.toml to the root .gitignore (tracked-modified). Run on a detached base checkout in CI (the Action fits on the PR base, then restores the head), those writes made the follow-up `git checkout` abort — the first PR to introduce argot.toml to a repo failed the non-blocking voice check at the fit step (get-tmonier/vigie#39). Scaffolding config is `argot init`'s job (moved there); fit now writes only inside the gitignored .argot/. config::load already returns the default in-memory when argot.toml is absent, so the fit stays correct. Plus defense-in-depth: the Action resets + cleans (without -x, preserving .argot/) before the restore checkout. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
damienmeur
added a commit
to get-tmonier/argot
that referenced
this pull request
Jul 13, 2026
… redundant display) (#114) * polish(landing): drop the monorepo min-numbers from the engine stats; punch up the demo verdict The two "2.3 min / 2.7 min" monorepo stats read as slow, not as a flex, on a landing hero — the machine-wide-cache story stays qualitatively in the fine print instead. The everyday-repo row (check/fit/refresh) is unchanged and stays a clean single row. Separately, the DemoTerminal cascade was a uniform metronome; tightened it (320→190ms) and made the verdict line land as the climax — it drops from higher with a back-ease overshoot while a stronger flag-flash ignites underneath. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * polish(demo): re-render the README GIF at 2× — crisp instead of blurry The old GIF rendered at 1280px / ~96 KB looked soft and banded when the README shows it at 760px. Bumped the VHS tape to 2000×760 at FontSize 34, 30fps, so it stays retina-crisp; render.sh handles the higher-res output. Same real `argot check --staged` scenario (a Django view in an all-FastAPI repo), re-rendered from the current binary. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(semantic): show the real function name in redundant/misplaced findings The semantic layer rewrites a function's own name to a neutral `f` before embedding (so a short function isn't dominated by its own identifier). That normalised text was stored as FunctionRef.text and leaked into the hunk a finding displays — users saw `def f(...)` instead of the code they wrote. Split the field: `text` is the real source (shown verbatim), `embed_text` is the name-normalised copy fed to the embedder and content-hash. Embeddings and cache keys are byte-identical (they still use the normalised text); only the displayed code changes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(fit): make argot fit side-effect-free on the tracked tree argot fit scaffolded a default argot.toml (untracked) and appended argot.local.toml to the root .gitignore (tracked-modified). Run on a detached base checkout in CI (the Action fits on the PR base, then restores the head), those writes made the follow-up `git checkout` abort — the first PR to introduce argot.toml to a repo failed the non-blocking voice check at the fit step (get-tmonier/vigie#39). Scaffolding config is `argot init`'s job (moved there); fit now writes only inside the gitignored .argot/. config::load already returns the default in-memory when argot.toml is absent, so the fit stays correct. Plus defense-in-depth: the Action resets + cleans (without -x, preserving .argot/) before the restore checkout. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * style: rustfmt pass Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(semantic): tolerate a NUL byte in source; surface the real embed error A source function can carry a raw NUL byte — e.g. a `\0` key separator inside a template literal (moneta's change-coordinator: `` `${a}\0${b}` ``). llama.cpp's tokenizer takes a C string and rejects an embedded NUL, so that one function failed to embed and — all-or-nothing — the whole language's semantic index failed to build, disabling the semantic layer for the repo. Strip NUL bytes before tokenizing (they carry no token; allocate only when one is present). Also surface the cause: the fit-time semantic error was printed with `{e}` (Display — only the outer "embed corpus functions" context), hiding the inner llama.cpp detail even under RUST_BACKTRACE. Print `{e:#}` for the full chain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
🎙️ argot voice check100% in-voice — this diff sounds like the rest of the repo. ✅
argot is a probabilistic guardrail — informational, it never blocks a merge. |
|
You are seeing this message because GitHub Code Scanning has recently been set up for this repository, or this pull request contains the workflow file for the Code Scanning tool. What Enabling Code Scanning Means:
For more information about GitHub Code Scanning, check out the documentation. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a non-blocking argot voice check to CI and commits the fitted voice config.
What's included
.github/workflows/argot.yml— runsget-tmonier/argot@mainon PRs intomain. Posts a sticky voice-score comment + inline SARIF code-scanning annotations. Never fails the build.argot.toml— committed voice config. Excludes peripheral packages (landing,video,tokens) so the voice reflects the shipped product (packages/app)..gitignore— ignore personalargot.local.tomloverrides.Notes