-
Notifications
You must be signed in to change notification settings - Fork 36
Audit candidate: Vincentwei1021/video-shotcraft #789
Copy link
Copy link
Closed
Labels
audit-completeAudit report generatedAudit report generatedcase-study-cleanAudit was clean (score >=90, security != BLOCKED) — fires auditor-exemplar.ymlAudit was clean (score >=90, security != BLOCKED) — fires auditor-exemplar.ymlexemplar-publishedExemplar artifact written to auditor/exemplars/<slug>.mdExemplar artifact written to auditor/exemplars/<slug>.mdprs-submittedPRs submitted to target repoPRs submitted to target repo
Description
Metadata
Metadata
Assignees
Labels
audit-completeAudit report generatedAudit report generatedcase-study-cleanAudit was clean (score >=90, security != BLOCKED) — fires auditor-exemplar.ymlAudit was clean (score >=90, security != BLOCKED) — fires auditor-exemplar.ymlexemplar-publishedExemplar artifact written to auditor/exemplars/<slug>.mdExemplar artifact written to auditor/exemplars/<slug>.mdprs-submittedPRs submitted to target repoPRs submitted to target repo
Repo
https://github.com/Vincentwei1021/video-shotcraft
Next Steps
audit-readyto trigger the audit workflowChecklist
echo719x2,tura-ai-agent,sarff0); no CONTRIBUTING.md / CLA gateWhy this was seeded by hand
auditor-discovercould not have found this repo. The artifact probe scored it 1 (.claude-plugin/plugin.jsonalone), below theMIN_ARTIFACTS=5floor, because no pattern anchored at a repo-rootSKILL.md— the canonical single-skill layout in the open Agent Skills spec. The probe is fixed in the same PR as this seed (now scores 2), but 2 is still under the floor: the repo's real NL surface is a 104-cardreferences/shots/tree that no probe pattern reaches. See the PR for that open design question.Pre-seed review (manual audit at
0022ec4)bin/nlpm-check.claude-plugin/plugin.json100,agents/openai.yaml100The audit workflow should re-derive all of this independently — the numbers above are from a hand pass over 3 artifacts and are not a substitute for the pipeline's own scorer, which will sample more of the corpus.
Expected to clear the exemplar gate (score >= 90, security != BLOCKED). Candidate rules to exemplify: R01 (zero vague quantifiers corpus-wide), R05 (205-line router over a 1,634-line reference layer), R07 (the 何时读哪个文件 dispatch table), R37 (104/104 reference pointers resolve; every numeric claim exact).