| name | post-github-validation-comment |
| description | Post a validation comment to a GitHub PR or issue using the standard format. Use when the user asks to post a comment to a PR or issue, post sandbox or ingest validation results, share validation findings on a PR or ticket, or anything similar that involves posting validation results as a GitHub comment. |
Post GitHub Validation Comment
This skill applies to both PRs and issues/tickets. The format is identical; only the target and posting command differ.
Pick the target
If the user named a PR or issue number, use it. Otherwise auto-detect the PR for the current branch:
gh pr list --head "$(git branch --show-current)" --json number,title --limit 1
Use the result. Only ask the user if the branch has no open PR or they reference a different target. If the target is an issue/ticket (not a PR), the user must provide the number — there is no auto-detect for issues.
Format
Each comment opens with the current datetime, followed by the setup command that produced the data being compared (sandbox load, ingest run, etc. — rename the heading to match), followed by one analysis block per result. Each analysis block is:
- Table with the headline numbers.
- Brief commentary — a sentence or two on what the table shows and what's caused by this PR vs. upstream drift.
- SQL in a
<details> block.
Datetime: Every comment must open with the current datetime on its own line, formatted as _YYYY-MM-DD HH:MM TZ_ (italic). Get it from date '+%F %H:%M %Z' rather than assuming — do not omit it.
Keep it brief. The tables are the deliverable. Commentary is one or two sentences per block, no more. Don't restate the numbers the table already shows, don't add a preamble before the date line, and don't pad with caveats or recaps. Let the data carry the comment.
Template:
_<YYYY-MM-DD HH:MM TZ>_
### Sandbox load command
```
python -m recidiviz.tools.load_views_to_sandbox \
--sandbox_dataset_prefix <prefix> \
auto <flags>
```
### <section title>
| <col> | <col> | <col> |
|---|---|---|
| … | … | … |
<one or two sentences of commentary>
<details><summary>Query</summary>
```sql
-- the SQL that produced the table above
```
</details>
Repeat the analysis block (table → commentary → SQL toggle) for every result. Do not improvise on the structure.
One comment per validation
Combine findings from the same setup run into one comment with multiple analysis blocks. Post separate comments only when the setup commands genuinely differ (e.g., a re-load with different sandbox flags or a new ingest after a fix).
Posting
Compose the body in a temp file so multi-line content survives shell quoting, then post with the command that matches the target type:
gh pr comment <PR#> --body-file /tmp/gh_validation_<num>.md
gh issue comment <issue#> --body-file /tmp/gh_validation_<num>.md
Return the comment URL gh prints.
To extend a previous comment instead of stacking a new one, add --edit-last — use it to consolidate; use a fresh comment when chronology matters.
No external IDs or PII
GitHub comments are externally visible. Never include external IDs (e.g. external_id, VSTH_ID, Contact_ID, raw state IDs) or PII (names, DOBs, addresses, free-text notes) in either the table or the SQL <details> block. Internal person_id values are fine. If a specific row matters, aggregate (counts/deltas) or replace the identifier with a placeholder; share specifics out-of-band if needed.
