| name | sf-loop-pruner |
| description | Recurring health watch for deployed bots. One invocation = one pass. Flags decaying / dead bots, can auto-pause them, persists a log so consecutive runs don't double-fire. Pair with the /loop skill for an always-on roster cleaner. Use when the user asks "watch my bots for edge decay automatically / auto-pause dying bots / kill dead edges on a loop / monitor roster health". |
sf-loop-pruner
The "always-on roster cleaner". Companion to sf-loop-hunter —
where the hunter looks outward for fresh edges, the pruner looks inward at
your deployed bots and pulls the dying ones.
State lives at ~/.claude/skills/sf-skills/pruner-state.json so a bot
that's been auto-paused once doesn't keep getting flagged on every loop.
Wrap with the /loop skill for continuous operation:
/loop 30m python SF-Skills/sf-loop-pruner/pruner.py --apply --include dead
What each run does
- Pull all ACTIVE bots.
- For each, fetch the underlying strategy and run the edge-decay heuristic
from
sf-edge-watch / sf_client.edge_health.
- Drop any bot already auto-paused in a prior run (from state file).
- Report decaying / dead bots. With
--apply, PATCH them to PAUSED and
record the action in state.
When to use
- "Auto-pause dying bots on a 30-minute loop"
- "Run a roster health check in the background during the stream"
- "Keep the bot list pruned without me babysitting"
How to run
# Dry-run, default thresholds (decaying + dead)
python sf-loop-pruner/pruner.py
# Auto-pause anything DEAD only (most conservative)
python sf-loop-pruner/pruner.py --include dead --apply
# Auto-pause decaying + dead
python sf-loop-pruner/pruner.py --include decaying,dead --apply
Safety
--apply respects the 10-second per-bot PATCH cooldown.- State file means a bot is only auto-paused once per loop history. To
re-arm, delete the state file or remove that bot's entry.
- Without
--apply, the script is read-only and safe to schedule
arbitrarily.
- This loop never creates bots; pair with
sf-rebalance or sf-deploy
to actually replace what's been pruned.
