The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user corrections stored in a local SQLite database to update instructions, creating a vector for indirect injection.
Ingestion points: Reads from ~/.pro-workflow/data.db (specifically learn-rule rows).
Boundary markers: None specified; instructions do not detail how external data is delimited before being processed by the optimizer LLM.
Capability inventory: The skill has the ability to overwrite local SKILL.md files and perform network operations via LLM provider APIs.
Sanitization: Not explicitly present; the skill relies on the LLM's "optimizer" and "evaluator" logic to filter content.
[COMMAND_EXECUTION]: The skill instructions include the execution of shell commands to interact with the local file system and databases.
Evidence: Use of /skill-optimize as a primary entry point and sqlite3 for inspecting optimization results in ~/.pro-workflow/data.db.
[DYNAMIC_EXECUTION]: The skill is designed for self-modification, which involves programmatically rewriting its own source instructions at runtime.
Evidence: The update and slow update stages generate and apply patches to the SKILL.md file based on LLM output.

skill-optimizer