openbrewerydb-data-quality-auditor
Audit OpenBreweryDB source CSV data quality and prepare issue-ready findings. Use when asked to inspect the openbrewerydb/openbrewerydb dataset for structural, identity, completeness, geospatial, or b
SKILL.mdskills/openbrewerydb-data-quality-auditor/SKILL.md
GitHubOpenBreweryDB Data Quality Auditor
Perform read-only audits of the source CSVs in a local openbrewerydb/openbrewerydb checkout. Produce an issue-ready report by default; never fix records, open implementation PRs, or write in the dataset repository.
Non-negotiable boundaries
- Treat the dataset checkout as read-only. Do not create reports, temporary files, branches, commits, or any other files inside it.
- Never edit CSVs or generated artifacts. Report proposed corrections rather than applying them.
- Never run
npm,npm install, an npm script, another package manager's equivalent, or an upstream script implementation. Do not reproduce an upstream mutating workflow. - Use only read-only inspection commands and the bundled stdlib helper. Put any explicitly requested saved output outside the dataset checkout.
- Do not open an issue or pull request and do not implement a fix. The deliverable is Markdown ready for the user or maintainer to paste into an issue.
Workflow
Ask for the local dataset root if it is unknown. Confirm it contains
data/; do not modify or sync the checkout.Select one or more modes:
structural,identity,completeness,geospatial, orstatus. Readreferences/audit-checks.mdfor definitions and triage rules.Run the helper from this skill directory, with the dataset path as an argument. It reads the live brewery type enum from
src/config.tsas text without executing upstream code. Markdown is the issue-ready default:python3 scripts/audit_dataset.py /path/to/openbrewerydb --mode structural --mode identity python3 scripts/audit_dataset.py /path/to/openbrewerydb --format jsonWith no
--mode, all modes run. The helper readsdata/**/*.csvrecursively and writes only to stdout.Review findings against the live CSV rows. Do not claim an automatic heuristic is a confirmed defect. Group related rows and retain file, row, ID, severity, and confidence.
For status findings, follow
references/status-verification.md. Status is not established by CSV values alone and requires current external evidence.Before finalizing each issue candidate, search open issues and pull requests in
openbrewerydb/openbrewerydbfor the ID, brewery name, source file, and defect type. Use read-only GitHub access, for example:gh issue list --repo openbrewerydb/openbrewerydb --state open --search '"<brewery name>" OR "<id>"' --limit 100 gh pr list --repo openbrewerydb/openbrewerydb --state open --search '"<brewery name>" OR "<id>"' --limit 100Inspect plausible matches with
gh issue vieworgh pr view. Ifghis unavailable, use GitHub web search and disclose the limitation if neither can be used. Link related items and mark duplicates or already-addressed findings rather than proposing a new issue.Format the result with
references/issue-report-template.md. Include scope, commands/modes, evidence, affected rows, related open issues/PRs, and a non-mutating suggested resolution. Separate confirmed defects from items needing maintainer review.
Reporting rules
- Prefer a small, evidence-backed issue over a bulk dump of weak signals.
- Use source-relative paths and one-based physical CSV row numbers.
- Treat missing IDs as review items because pending contributor rows may intentionally lack IDs.
- Never infer that
(0, 0), a duplicate-looking identity, or an old web listing is conclusively wrong without corroboration. - State that no dataset files were changed and no npm or upstream implementation was run.
References
- Audit definitions and severity guidance:
references/audit-checks.md - Issue-ready output structure:
references/issue-report-template.md - Status evidence and conflict handling:
references/status-verification.md