The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a potential for Indirect Prompt Injection (Category 8) because it automatically incorporates data from a local database into the agent's core instructions.
Ingestion points: The skill reads "learn-rule" rows (accumulated corrections) from the SQLite database located at ~/.pro-workflow/data.db.
Boundary markers: There are no specified delimiters or instructions to ignore embedded malicious content within the corrections data being processed.
Capability inventory: The skill has the capability to overwrite SKILL.md files, which define the agent's operational logic and safety constraints.
Sanitization: The skill does not appear to perform sanitization or validation of the correction data before using an LLM to generate instruction patches.
[COMMAND_EXECUTION]: The skill instructions and architecture describe a workflow that involves executing shell commands and database queries.
The documentation references using sqlite3 to query the internal ~/.pro-workflow/data.db database.
The skill is invoked via a command-line style interface (/skill-optimize) which executes a multi-stage pipeline including file modifications and database reads.

skill-optimizer