GitHub Org Chart

Produces two artifacts in ${TMPDIR:-/tmp}/gh-org-chart/:

• <org>-org.json — canonical data: teams, members (with role), repos (with archived flag and permission), CODEOWNERS paths. Cache + hand-editable source of truth.
• <org>-org.html — single self-contained interactive explorer rendered from the JSON. Opens via file://, works offline.

Set OUT_DIR once at the start of every phase and reuse it:

OUT_DIR="${TMPDIR:-/tmp}/gh-org-chart"
mkdir -p "$OUT_DIR"

Arguments

• /gh-org-chart — prompt for org.
• /gh-org-chart <org> — render-if-fresh: if <org>-org.json exists and mtime is within 24h, re-render from it. Otherwise collect, then render.
• /gh-org-chart <org> --refresh — force re-collect.
• /gh-org-chart <org> --no-codeowners — skip CODEOWNERS scan (faster on big orgs).
• /gh-org-chart <org> --no-members — drop members from collection and output.

Phase 1: Intake

1. Resolve org: if provided as argument, use it. Otherwise ask:

   Which GitHub org should I chart?

2. Verify auth: run gh auth status. Confirm read:org is in the scopes line. If not, instruct:

   gh auth refresh -s read:org then re-run the skill.

Phase 2: Decide collect vs. reuse

3. Locate $OUT_DIR/<org>-org.json.

4. If --refresh was passed, or the file does not exist, or its mtime is older than 24h, run collect (Phase 3). Otherwise skip to Phase 4.

   Freshness check:

   JSON="$OUT_DIR/$ORG-org.json"
   if [[ "$REFRESH" == "1" ]] || [[ ! -f "$JSON" ]] || \
      [[ $(($(date +%s) - $(stat -f %m "$JSON" 2>/dev/null || stat -c %Y "$JSON"))) -gt 86400 ]]; then
     NEEDS_COLLECT=1
   fi

Phase 3: Collect

5. Run the bundled collect script (handles zsh noclobber via rm -f):

   rm -f "$OUT_DIR/$ORG-org.json"
   "${CLAUDE_SKILL_DIR:-$HOME/.claude/skills/gh-org-chart}/scripts/collect.sh" \
     "$ORG" $FLAGS > "$OUT_DIR/$ORG-org.json"

   Where $FLAGS is built from --no-codeowners and --no-members if set. CODEOWNERS scanning is the long pole — expect ~1 API call per owned repo. Big orgs (200+ owned repos with rate limiting): consider --no-codeowners.

Phase 4: Render

6. Run the renderer:

   python3 "${CLAUDE_SKILL_DIR:-$HOME/.claude/skills/gh-org-chart}/scripts/render.py" \
     "$OUT_DIR/$ORG-org.json"

   This writes <org>-org.html next to the JSON (i.e. in $OUT_DIR).

7. Open it (macOS):

   open "$OUT_DIR/$ORG-org.html"

   Other platforms: report the path so the user can open it themselves.

Phase 5: Report

8. Summarize (set JSON="$OUT_DIR/$ORG-org.json"):
   • Teams: jq '.teams | length' "$JSON"
   • Owned repos (admin or maintain): jq '[.teams[].repos[] | select(.permission == "admin" or .permission == "maintain") | .name] | unique | length' "$JSON"
   • CODEOWNERS path attributions: jq '[.teams[].repos[].codeowner_paths // [] | length] | add // 0' "$JSON"
   • Member entries: jq '[.teams[].members[]] | length' "$JSON" (omit if --no-members).
   • Whether the JSON was freshly collected or reused from cache.

Notes

• Hand-editing the JSON: edits survive across /gh-org-chart <org> runs because the freshness check reuses the file. Use --refresh when you want collection to overwrite your edits.
• CODEOWNERS attribution: only @<org>/<team-slug> owners produce attributions. Individual user owners (@alice) and external orgs (@other-org/team-x) are intentionally ignored — this is a team-ownership view.
• No reporting lines: GitHub teams reflect permission grouping, not management hierarchy. Manager → report relationships need a different data source (HRIS).
• Performance: CODEOWNERS scan is restricted to owned repos (permission ≥ maintain) to keep API calls bounded. Collecting roles adds one ?role=maintainer call per team.
• Data shape (for hand-editing / jq): members is [{login, role}] where role is "maintainer" or "member"; each repo carries permission and an archived flag. --no-members leaves members empty.