agentleFS
dork-labsdork-labs/dorkos/.agents/skills/deps-weekly/SKILL.md

deps-weekly

Skill11 starsChanged 50 days agoMITView on GitHub

Monday DorkOS dependency upkeep, called from @dorkos-qa's heartbeat: runs /app:upgrade unattended in a throwaway worktree, lands one PR per lane, files tickets for majors, and posts one line to @dorkos-pm. Never bumps the runtime SDKs (that is runtime-watch). Use only when a heartbeat checklist or a person asks for the weekly dependency upkeep.

  • Installs packages

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Clean up: Remove $TMPDIR/deps-weekly and any lane worktree whose PR is open and pushed.

    line 24
  2. Distinctive:

    Run paid evals, set a paid flag, or spend money in any form.

    line 37

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

deps-weekly

The weekly dependency pass. It is /app:upgrade in its default auto mode, wrapped in the rules that make it safe to run from a heartbeat with nobody watching. It is a skill, not a schedule: @dorkos-qa's heartbeat calls it on Monday, so it runs with that agent's permissions and raises no approval cards.

The procedure is .claude/commands/app/upgrade.md; the judgment is the maintaining-dependencies skill (.claude/skills/maintaining-dependencies/SKILL.md). Read both before step 3. This file adds only what a heartbeat run needs on top.

Run

  1. Work in a throwaway copy, never a shared checkout. git fetch origin, pin BASE=$(git rev-parse origin/main) and compare against $BASE for the whole run. Clear a copy a killed run left behind (git worktree remove --force "$TMPDIR/deps-weekly" 2>/dev/null; git worktree prune), then git worktree add --detach "$TMPDIR/deps-weekly" "$BASE" and run everything from there. Each lane branches from it.
  2. Check room first. Count worktrees on this machine that hold a node_modules. If two already do, stop and report "skipped: disk" rather than add a third. Check df -h /: under 20 GB free, stop and say so. Then pnpm install --frozen-lockfile --prefer-offline and pnpm --filter @dorkos/shared build, on Node 24.
  3. Run the procedure in .claude/commands/app/upgrade.md with no arguments (auto). In Claude Code that is /app:upgrade; in any other harness, read the file and follow it step by step. Its autonomous contract holds as written: one PR per lane, idempotent branch prefixes, the ledger edited with the override, nothing merged by hand.
  4. Apply the DorkOS landing rules to every PR the run opens:
    • Commits are chore(deps), never fix( or refactor(, so no changelog fragment is seeded. Label skip-changelog and review:light.
    • A PR that touches auth, permissions, connectors, mini apps (extensions), relay or Cloud auth opens with the hold label and stays unarmed. Post it in #dorkos for @security-auditor; only their OK there lifts it.
    • Arm your own green PR with gh pr merge --auto <n> only when it needs no Security review. Never an admin merge, never an empty commit, never a branch update to satisfy a gate.
    • Patches and security fixes land on green. A major, or anything the run calls breaking, gets a ticket for @dorkos-pm and no commit.
  5. Leave the runtime SDKs alone. The families in .claude/config/runtime-deps.json, plus the Pi packages in packages/doe, are excluded by the procedure's rule 2. If the run sees them drift, it names them in the report and points at runtime-watch.
  6. Escalate security by severity. A critical or high advisory that the run fixed: ask @dorkos-pm for a same-day patch release once its PR merges. Critical with no fix: a PR (or a ticket with the blocker) within 48 hours. High: within 7 days. Medium and low: the monthly batch.
  7. Clean up. Remove $TMPDIR/deps-weekly and any lane worktree whose PR is open and pushed. Spare nothing that holds unpushed work.

Report

One line to @dorkos-pm, in #dorkos, in this shape:

Deps (Mon <date>): updated <n> (<PR titles>), waiting <n> (<why>), alerts crit/high/med/low <a>/<b>/<c>/<d>, runtimes: <see runtime-watch | none>.

A quiet week is one line too: Deps (Mon <date>): nothing to update, alerts 0 open. A lane that could not finish is the first thing in the line, never left out.

Never

  • Bump a runtime SDK, a major, or a .github/dependabot.yml family here.
  • Run paid evals, set a paid flag, or spend money in any form.
  • Touch another agent's worktree or PR, git stash, git checkout -- <path>, or kill a process by name.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.