The Agent Skills Directory

[PERSISTENCE]: The skill configures a daily automated task using the clawdbot cron add command, which establishes a persistent execution schedule on the host system.
[COMMAND_EXECUTION]: The skill performs extensive shell command execution to detect installation types and update software, including calls to npm, pnpm, bun, and clawdbot.
[EXTERNAL_DOWNLOADS]: The updater fetches code from external sources, specifically global package registries and the ClawdHub registry, to update the agent and its extensions.
[PRIVILEGE_ESCALATION]: Documentation in references/agent-guide.md explicitly advises the agent to suggest sudo to the user if permission errors (EACCES) are encountered during global updates.
[DYNAMIC_EXECUTION]: The agent is instructed to write a multi-line bash script to a local directory (~/.clawdbot/scripts/auto-update.sh) and then schedule its execution, representing the creation and running of executable content at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill creates a potential attack surface by capturing the full output of external update commands and requesting the agent to parse and summarize it. Ingestion points: captured output of clawdhub update --all. Boundary markers: none. Capability inventory: includes global command execution (npm update) and system modifications (cron add). Sanitization: none.
[TIME_DELAYED_CONDITIONAL]: Execution of update logic is deferred and triggered by a time-based condition via the system's cron scheduler, which can be used to hide malicious activity until a specific time.

auto-updater