clawdbot-backup
Fail
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEPERSISTENCEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill explicitly targets sensitive files for backup and synchronization that are known to store authentication tokens and API keys.
- Evidence: The backup commands and the
clawdbot-backup.shscript specifically target~/.claude/settings.jsonand~/.claude/mcp/servers.json(MCP configurations). - [PERSISTENCE]: The skill provides comprehensive instructions and pre-written configuration files to establish persistent execution of scripts across system reboots and user sessions.
- Evidence: Instructions for creating daily cron jobs (
crontab -e), systemd user services and timers (clawdbot-backup.service,clawdbot-backup.timer), and macOS Launch Agents (com.clawdbot.backup.plist). - [DATA_EXFILTRATION]: Several methods provided for syncing and backing up data involve transmitting the contents of the sensitive
~/.claudedirectory to external locations. - Evidence: Instructions to use
git pushto remote repositories (e.g.,git remote add origin git@github.com:username/clawdbot-config.git) andrsyncto sync data to remote servers (user@server:~/clawdbot-backup/). - [COMMAND_EXECUTION]: The skill contains multiple shell scripts and complex command strings that perform file system modifications and network operations.
- Evidence: The restoration commands use
tar -xzvfwhich can overwrite arbitrary files in the user's home directory (-C "$HOME"). Thecleanup_backupsfunction in the provided script usesxargs rm -vto delete files. - [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by ingesting untrusted data from the local file system and using it in powerful shell operations.
- Ingestion points:
~/.claude/skills/(containingSKILL.mdfiles) and~/.claude/settings.jsoninSKILL.md. - Boundary markers: None present.
- Capability inventory: Shell execution for
tar,git,rsync, andrmacross multiple scripts. - Sanitization: None present. The skill does not sanitize the contents of the files it backs up, allowing instructions embedded within those files to potentially influence the agent's behavior during the backup or restore process.
Recommendations
- AI detected serious security threats
Audit Metadata