The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use sqlite3 for reading and writing to the local Digikam database (~/Pictures/digikam4.db). It also utilizes shell commands such as cp to create database backups and cat to retrieve metadata from local files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing untrusted metadata from external YAML files to drive its classification logic.
Ingestion points: The agent ingests data from uploader, title, and description fields in YAML files located at /Users/mtm/pdev/taylormonacelli/grouchygiraffe/data/.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious instructions embedded within the YAML metadata content.
Capability inventory: The agent is authorized to execute sqlite3 commands to modify database records and use git to commit progress notes to an Obsidian vault.
Sanitization: No sanitization or input validation is specified for the metadata before it is processed by the agent's logic. This is mitigated by a mandatory human-in-the-loop confirmation step before any database modifications occur.

moonlitlynx-food-run