Swarm Janitor

Enterprise-grade OpenClaw skill for cleaning up orphaned subagent processes, archiving transcripts to SuperMemory, and freeing disk space without losing work. Features dry-run mode, configurable retention policies, and comprehensive safety checks.

Do not run the shell/python commands shown in SKILL.md until you have the actual implementation to inspect. Specific steps to take before installing or enabling automation: - Verify origin: visit the claimed homepage (https://github.com/openclawdad/swarm-janitor) and confirm the repository contains the scripts (scripts/swarm_janitor.py) and matches the SKILL.md. - Inspect the code: review the actual Python scripts for exactly how they detect processes, determine 'orphaned' sessions, and delete files. Look for any network calls, unexpected endpoints, or credential exfiltration. - Confirm env and creds: SUPERMEMORY_API_KEY (and any S3 or other credentials) are referenced in config.yaml but not declared; understand where and how those values are used before supplying them. - Test safely: run any discovered script in a safe test environment or a copy of your sessions directory with dry-run enabled. Do not run --clean --force on production until you’ve confirmed behavior. - Check permissions: cron example writes to /var/log and may require root or elevated permissions; consider running under a restricted account and ensure proper backups exist. - If you cannot find the implementation repo or the scripts referenced, treat the package as incomplete/untrustworthy and do not enable automated deletions. Because the package is missing the executable scripts and has undeclared env and path implications, proceed cautiously; the inconsistencies could be benign (incomplete packaging) but could also hide risky operations if the missing code is later obtained from an untrusted source.

Type: OpenClaw Skill Name: swarm-janitor Version: 0.1.0 The OpenClaw AgentSkills skill 'swarm-janitor' appears benign. Its stated purpose is to clean up orphaned subagent processes and archive transcripts, which involves file system access (deleting sessions, writing logs) and potential network access (archiving to SuperMemory, optional metrics webhook). All described actions and configurations in SKILL.md and references/config.yaml align with this legitimate maintenance function. There is no evidence of prompt injection, malicious execution, or intentional data exfiltration; the optional `webhook_url` for metrics in `references/config.yaml` is set to `null` by default, and SuperMemory archival is a core, stated feature.