agentleFS
VorElzeVorElze/hermes-office-skills/skills/hermes-profile-hygiene/SKILL.md

hermes-profile-hygiene

Skill0 starsChanged 20 days agoMITView on GitHub

Use when cleaning finished Kanban cards and dead jobs.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 1 is distinctive, and 7 are 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.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After a project closes, before onboarding the next one.

    line 23
  2. Very specific:

    Only ✓ done cards qualify.

    line 44
  3. Very specific:

    Verify each with a read-back (is-active → inactive, is-enabled → disabled, no listener, vhost symlink absent) and state the distinction: …

    line 71
  4. Very specific:

    Prefer hermes cron pause <id> over hermes cron remove <id>: pause stops the quota burn but stays reversible for a revision round.

    line 86
  5. Very specific:

    no reference in cron/jobs.json (grep the live config, not logs/history),

    line 97
  6. Very specific:

    no reference in config.yaml or other live config,

    line 98
  7. Very specific:

    no running process (pgrep -af <name> returns nothing).

    line 99
  8. Very specific:

    Never truncate active runtime logs (agent.log, gateway.log, errors.log) — they are the diagnostic trail for live incidents.

    line 111

≈ “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

Hermes profile hygiene

Clear finished cards and dead artifacts from a profile so the board and job list show only live work — without touching anything a worker is actively using. The goal is a board where every visible card is a real next task.

When to Use

  • "Bersihkan sampah / kartu yang sudah terpakai / perintah menumpuk" — the board is buried in done cards and the cron list confuses new work.
  • After a project closes, before onboarding the next one.
  • The operator reports difficulty finding current work among old results.

Procedure

1. Inventory and classify before deleting anything

Batch these reads in one turn:

hermes kanban boards list       # counts per status per board
hermes kanban list              # card lines: ✓ done, ◻ todo, ● running
hermes cron list --all          # every job including paused/disabled

Classify each artifact into: live (leave), finished (archive/prune), orphan (candidate for deletion after verification in step 4). Report the inventory before mutating — the operator should see the same numbers you do.

2. Cards: archive done cards, never purge

  • Only ✓ done cards qualify. todo/running/blocked cards are live work even when their titles look obsolete — a decomposer chain (parent + auto- children) is one unit; if any member is active, leave the whole chain.
  • Confirm liveness before touching non-done cards: hermes kanban show <id> and look for recent heartbeat/claim timestamps (minutes-old = running now).
  • Archive in one batch, ids parsed from the ✓ lines of hermes kanban list:
hermes kanban archive t_id1 t_id2 ...   # hides from board, history kept
hermes kanban gc                        # prunes events/logs past retention
  • archive keeps comments and run history readable via hermes kanban show — that is the point: out of sight, still auditable. archive --rm is irreversible; only run it when the operator explicitly asks for purge.
  • gc clears task events and worker logs (30-day defaults), not cards.

3. Services and routes: decommission as a unit

A project service usually has three coupling points; all three must be handled in one pass or it resurrects or leaves error paths:

systemctl --user stop <svc>.service && systemctl --user disable <svc>.service
ss -ltnp | grep ':<port>'          # confirm the listener is gone
a2dissite <vhost> && apache2ctl configtest && systemctl reload apache2  # drop reverse-proxy route
  • Verify each with a read-back (is-active → inactive, is-enabled → disabled, no listener, vhost symlink absent) and state the distinction: stopped/disabled (recoverable, survives reboot) vs deleted (source removed) — report exactly which was done, never blur them.
  • Leaving the proxy vhost in place after the backend dies is what turns a deliberate shutdown into recurring 502s for anyone hitting the URL.
  • Preserve source/config backups when decommissioning unless deletion was explicitly requested; removing test scripts, probe scripts, and session login files belonging to the project is part of the pass.

4. Cron: justify every job; pause instead of delete when unsure

  • Walk hermes cron list --all and give each job a one-line reason it still earns its keep (standing absensi reports, credential watchdogs, guards). Jobs with no reason and no owner project are deletion candidates.
  • Prefer hermes cron pause <id> over hermes cron remove <id>: pause stops the quota burn but stays reversible for a revision round; remove is for jobs whose project is verifiably closed AND the operator approved.
  • Edit jobs through the CLI (hermes cron edit), never by hand-editing cron/jobs.json under a live gateway.

5. Scripts and backups: verify orphanhood on multiple signals

A filename that looks stale proves nothing. A script is an orphan only when ALL hold:

  1. no reference in cron/jobs.json (grep the live config, not logs/history),
  2. no reference in config.yaml or other live config,
  3. no running process (pgrep -af <name> returns nothing).

Then delete. Same rule for jobs.json.bak-*: keep the newest backup, drop older ones — one recovery point is enough.

6. Caches and logs

  • cache/scratch/*: files idle ≥24h are auto-pruned by the scheduled prune cycle — do not force-delete merely because they are old; act only under disk pressure or for obviously dead artifacts (checked with ls -la).
  • cache/terminal-output/*: session command logs, safe to delete past a few days.
  • Never truncate active runtime logs (agent.log, gateway.log, errors.log) — they are the diagnostic trail for live incidents.

7. Verify and report

Re-run hermes kanban boards list and hermes cron list --all; the board should show only live cards and the cron list only justified jobs. Report with measured numbers: cards archived, files deleted, MB freed, jobs removed/paused — plus an explicit "what was NOT touched and why" line for anything live. This matches the operator's standing expectation of evidence over assertions.

Pitfalls

  • Don't judge card liveness by status or title alone — a running card with heartbeats seconds ago is live even if its task sounds finished, and a todo card can be a child of an active chain. Check show timestamps.
  • Don't archive --rm during routine cleanup — purge destroys the audit trail (comments, run evidence) the operator uses to reconstruct decisions; archive achieves the clutter goal with zero data loss.
  • Don't delete a script on filename intuition — the reference may live in a cron job or a live card; grep the live configs and pgrep first.
  • Don't treat gc reporting 0 rows as "nothing to clean" — gc covers events/logs only; card clutter is a separate archive action.
  • Don't delete cron jobs just because they are quiet — quiet standing jobs (weekly guards, hourly watchdogs) may fire rarely by design; classify by ownership, not activity.
  • Don't skip the verify pass — archiving in batch can leave stragglers (cards that changed status mid-run), and the final counts are the proof the operator asked for.

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.