The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions to establish a monthly automated maintenance task. This configures the agent to execute specific actions on a recurring schedule, creating persistent automated behavior in the environment.
Evidence: The 'Monthly Maintenance Cron' section in SKILL.md provides a snippet for cron action=add with a monthly schedule.
[COMMAND_EXECUTION]: The skill includes a direct instruction for the agent to execute a destructive filesystem command to delete a specific directory.
Evidence: The command rm -rf ~/.clawdbot/memory/lancedb/ is provided in SKILL.md as the method for wiping memory.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow for reseeding memory involves parsing an external file (MEMORY.md) and storing its contents as facts. This ingestion of untrusted data into the agent's vector memory creates a risk of instructions hidden in the text being executed or influencing future agent behavior.
Ingestion points: MEMORY.md is identified as the source file to be parsed and stored in LanceDB.
Boundary markers: Absent. There are no instructions to use delimiters or to treat the file content as purely passive data.
Capability inventory: The skill has the ability to remove files (rm -rf), modify agent configuration (gateway action=config.patch), create scheduled jobs (cron action=add), and write to vector memory (memory_store).
Sanitization: Absent. No mention of filtering, escaping, or validating the content of the MEMORY.md file before it is processed.

memory-hygiene