Repository navigation
[SCR-214] Roll the label vocabulary - #215
Conversation
The issue standard classifies with labels and the labels existed in one repo out of five. An audit of all 133 open issues reported every one missing a route and a state label, and for four of the five repos that was not the author being sloppy. The labels did not exist to apply. The full rollout could not fix it. It plans a branch name, so it skips a repo with no .github/pr-standards.json, and that is 55 of 133 repos, most of the fleet. Labels need no branch and no prefix. --labels-only skips only a repo whose issues are disabled, so it reaches everything the full mode cannot, including a repo that has already adopted the forms and simply wants the vocabulary kept current. The tally counts repos in both modes. Saying labels there would report 13 labels across one repo as one label.
|
Warning Review limit reachedNext included review available in 48 minutes. View limit detailsLimit details: You’ve used the included review currently available. Your 137 included PR review attempts over the past 7 days set your current allowance at 1 review per hour. Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Team Run ID: 📒 Files selected for processing (1)
Comment |
There was a problem hiding this comment.
Reviewed head: c4c8e84b497301429b62790f6775f894517d8fd4
No fixes pushed — every council claim was checked against the surrounding code and did not hold up.
Checked and discarded:
- "A second
--labels-only --applyrun still prints 'Labels applied where planned' instead of a no-change result" (correctness + scope lenses). The tally message reflects the desired end state ("labels are as planned"), not a before/after diff — that's the same behavior the pre-existing full-mode path has always had for its own label loop (gh label create --forcemakes the state a no-op; the summary text was never wired to detect "nothing changed"). Not a regression introduced by this diff, and no acceptance criterion in the PR or repo requires distinct no-change wording. The PR's own proof ("rerun 14/14, no change") is describing label state viacomm, not the script's printed line. - "13 vs 14 label count mismatch between code comment and proof". Not a mismatch: the script deliberately creates 13 labels and skips
bug(line 54-55: "bugis omitted: GitHub creates it on every repo, and--forcewould rewrite a description someone may have changed"). 13 script-managed + 1 pre-existingbug= the 14/14 the proof counts. Consistent. - "No proof of
--labels-onlyagainst an issues-disabled repo". Thehas_issuesskip (lines 117-125) is shared, pre-existing logic this diff does not touch — both modes hit it identically before and after this change, so it isn't a code path this PR introduces or modifies, and doesn't need fresh proof.
I also independently checked: shell-injection surface (label name/color/description come from the static LABELS heredoc, not attacker input — no risk), set -u safety of the new adopted/prefix variables (only read inside the same branch that sets them), single-commit branch matching the reviewed SHA, and PR-standard compliance ([SCR-214] Roll the label vocabulary, one Closes #214, ## What/## Why/## How I verified, Assisted-by: trailer, 51+29 lines / 1 file — well under the 500-line/40-file caps).
One pre-existing (not diff-introduced) wart worth a human's attention someday: --help has always truncated its own last sentence ("...one issue and one pull request" cuts off "per repo, each under the caps." — true both before and after this diff, since the sed range was shifted by exactly the 3 lines this diff added). Not raised as a finding since this PR didn't create it and didn't touch that sentence.
🧑⚖️ Council
| Model | Lens | Result |
|---|---|---|
| GPT-5.6 (Codex) (via OpenRouter) | correctness | rejected |
| Gemini 3 Pro | performance | no findings |
| Kimi K3 (via OpenRouter) | security | no findings |
| Grok 4.5 | maintainability | no findings |
| GPT-5.6 (scope) | scope | rejected |
Closes #214
What
issue-standards-rollout --labels-only: creates the 13-label vocabulary and plans nothing else. Dry by default like the rest of the program.Why
The issue standard classifies with labels, and the labels existed in one repo out of five. GitHub issue types are an organisation feature and this account is a user, so labels are the only classification axis there is.
An audit of all 133 open issues reported every one of them missing a route and a state label. For four of the five repos that is not the author being sloppy: the labels did not exist to apply.
The full rollout could not fix it, because it plans a branch name and therefore skips any repo with no
.github/pr-standards.json. That is 55 of 133 repos, most of the fleet. Labels need no branch and no prefix, so--labels-onlyreaches all of them, and also reaches a repo that already adopted the forms and simply wants the vocabulary kept current.How I verified
It reaches the repos the full mode skips, which is the entire point:
Default mode is unchanged, and dry mode still writes nothing:
Applied for real against
pooriaarab/skills, counted withcommagainst the expected set rather than a hand-rolled filter:The first count I wrote for this was wrong and worth naming:
map(select(LIST|index(.)))drops the first element of the list, becauseindexreturns0for it and0is falsy in jq. It reported9/13before and22/13after, and 22 of 13 is what made it obvious. The numbers above come fromcommon two sorted lists.On idempotency
gh label create --forceis what makes a rerun a no-op, and the code now says so where it happens rather than leaving it implied.How it was built
Delegated to
grok-4.6-xhighthrough thedevinCLI. The mode, the skip logic and the comments explaining why labels need no prefix are its work, and it left the default path untouched, which was the main risk.Review found one defect: the summary took its unit from the mode, so
--labels-onlyprinted1 label(s) consideredwhile$countwas counting repos. That was my brief's fault rather than the model's, since I asked for exactly that wording. It now reads1 repo(s), labels only.Assisted-by: claude-personal-1:claude-opus-5