ghreport

Files (2)

• SKILL.md
• ghreport.sh · scripts/

SKILL.md (first lines)

# ghreport: what state are an org's repos in?

`ghsync` answers "do I have all the code locally and up to date?" `ghreport`
answers the companion question: "across everything I can see, what needs
attention right now?" It is **read-only** - it never clones, pulls, or mutates
anything.

It reuses `ghsync`'s discovery so the team-union / org-repo-list logic lives in
exactly one place: `ghreport` shells out to `ghsync --porcelain` to get the
deduplicated repo list, then queries each repo's remote state over `gh api`.

## What it reports, per repo

| Signal | Source | Reported as |
|--------|--------|-------------|
| Open PRs | `repos/{o}/{r}/pulls?state=open` | count (drafts noted) |
| CI on default branch | latest completed run per workflow on the default branch | `pass` / `fail` (names the failing workflows) / `none` |
| Security alerts | Dependabot + code-scanning + secret-scanning, `?state=open` | three counts, or `n/a` (no-access) |
| Branch protection | `repos/{o}/{r}/branches/{default}/protection` | `protected` / `unprotected` / `unknown` |

### The no-access rule

Security-alert and branch-protection endpoints require admin on the repo. When
the caller is denied (`403`/`404`), `ghreport` reports `n/a` (alerts) or
`unknown` (protection) - **never** a clean `0`. Reporting "no access" as "no
problems" would be a dangerous lie, so the two are kept distinct. A repo whose
own metadata can't be read at all is listed as `could not assess`.

## How to run it

The script lives next to this file. Claude Code writes this skill's directory
into the script paths below when it loads the skill, so each path is already
absolute:

```bash
# Org is derived from the directory you run in, exactly like ghsync. To report
# on the "meridianhub" org, run from a directory named meridianhub:
cd ~/dev/github.com/meridianhub
bash "${CLAUDE_SKILL_DIR}/scripts/ghreport.sh"
```

Override the org or target directory explicitly when they differ:

```bash
bash "${CLAUDE_SKILL_DIR}/scripts/ghreport.sh" --org meridianhub --root ~/dev/github.com/meridianhub
```

## What to do when invoked

1. **Run it** from (or pointed at) the org directory. It streams a progress dot
   per repo, then prints the summary.
2. **Read the headline** - the counts of repos with failing CI, open alerts,
   and an unprotected default branch. These are the repos that need attention.
3. **Surface the "Needs attention" list** to the user verbatim; that is the
   actionable core of the report.
4. **Point at the saved file** (`org-state-<org>-<date>.md` in the root) for the
   full per-repo table.

## Flags

| Flag | Effect |
|------|--------|
| `--org NAME` | Org or personal account to report on (default: basename of `--root` / cwd) |
| `--root DIR` | Org root; also where the report file is written (default: cwd) |
| `--limit N` | Only the first N repos (quick test against a large org) |
| `--quiet` | Suppress the per-repo progress dots; keep the summary and file |
| `--no-file` | Terminal only; do not write the markdown file |
| `--render-dir DIR` | Skip discovery + querying and render a report from a directory of previously-collected per-repo JSON fragments (replay / testing) |

## Output

- **Terminal:** a one-line headline (counts that tripped a signal), a rollup
  table, and a "Needs attention" section listing only the repos with an issue.
- **File** (unless `--no-file`): `org-state-<org>-<YYYY-MM-DD>.md` in the root,
  with the full table and per-section detail.

## Requirements

- **`gh` and `jq`** on `PATH`, and `gh auth status` authenticated (the same
  prerequisites as `ghsync`; discovery is delegated to it).
- For a **GitHub Enterprise** host, set `GH_HOST` or `gh auth login --hostname`
  first, as with `ghsync`.
- Very large orgs make many REST calls (~5 per repo); use `--limit` to sample.

From bjcoombs/ai-native-toolkit, Apache-2.0.

More from bjcoombs/ai-native-toolkit