running-tend

name	running-tend
description	Tend-specific guidance for tend CI workflows. Adds non-standard workflow inclusion for usage analysis and repo conventions on top of the generic tend-* skills.
metadata	{"internal":true}

Tend CI

Repo-specific guidance for tend workflows running on tend itself. The generic skills (tend-running-in-ci, tend-review, tend-triage, etc.) provide the workflow framework; this skill adds tend conventions.

Filing issues in other repos

Standing exception granted: file directly in agent-equipped targets without asking permission here first. Most tend consumers in data/consumers.json qualify, as do other Claude-Code-action-using repos. The default rule (open an issue here asking permission first) still applies when the target shows no agent signals.

Non-standard workflows

Tend has Claude-powered workflows beyond the generated tend-* set:

Workflow	File	Schedule	Purpose
review-reviewers	review-reviewers.yaml	47 * * * *	Hourly analysis of adopter repo sessions

These use the tend composite action and produce claude-session-logs* artifacts, but their names don't match the tend-* prefix that scripts filter on by default.

Usage analysis

Pass extra prefixes when running token reports or listing runs so these workflows are included:

"${CLAUDE_PLUGIN_ROOT}/scripts/token-report.sh" 24 "review-"
TARGET_REPO=max-sixty/tend "${CLAUDE_PLUGIN_ROOT}/scripts/list-recent-runs.sh" "tend-" "review-"

Labels

• claude-behavior — findings from review-reviewers
• review-runs — findings from review-runs

Session Log Paths

Artifact paths: -home-runner-work-tend-tend/<session-id>.jsonl

review-reviewers runs produce one session log per matrix repo in .github/workflows/review-reviewers.yaml.

Weekly: refresh data/consumers.json

Public repos that have installed tend. Read by the website's data Worker (see docs/website-data.md) to power the currently-tending dot, activity feed, and stat strip. Needs no opt-in because the workflow files are public.

# 1. Discover consumer repos via code search. Generated workflows pin a
#    version tag (`max-sixty/tend@X.Y.Z`, or `/codex@X.Y.Z`), so search the
#    bare `max-sixty/tend` token (version-agnostic; GitHub code search does
#    not index `@` or `/`, so this matches both the Claude and Codex refs).
#    The `.github/workflows/tend-` path filter below bounds precision.
mapfile -t REPOS < <(
  gh search code 'max-sixty/tend' --limit 100 --json repository,path \
    | jq -r '.[] | select(.path | startswith(".github/workflows/tend-")) | .repository.nameWithOwner' \
    | sort -u
)

# 2. Resolve bot_name from each repo's .config/tend.yaml.
mkdir -p data
{
  for repo in "${REPOS[@]}"; do
    bot=$(gh api "repos/$repo/contents/.config/tend.yaml" --jq '.content' 2>/dev/null \
      | base64 -d 2>/dev/null \
      | yq '.bot_name // ""' 2>/dev/null)
    [ -n "$bot" ] || continue
    jq -nc --arg repo "$repo" --arg bot "$bot" '{repo: $repo, bot_name: $bot}'
  done
} | jq -s . > data/consumers.json

Open a PR titled chore: refresh consumers.json if the file changed. Skip the PR (no diff to land) when git status --porcelain data/consumers.json is empty — git diff --quiet returns 0 for untracked paths, so the first-run case would no-op. Code search is 10 req/min — one call covers the whole list.

Weekly: integration test

End-to-end check that a fresh install completes and the generated workflows respond to a real issue and PR. Open references/integration-test.md and follow the recipe in order; do not skip the cleanup step even on assertion failure.