The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Python scripts (skillrules.py, review.py, and taste.py) using python3 or py. These commands take arguments such as <rule>, <reason>, and <name> which are populated from external data sources.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from edits/<name>/review.json and edits/<name>/corrections.jsonl (comments and corrections from video reviews). This content is used to generate the strings passed to the shell commands.
Ingestion points: Video review notes (review.json) and correction logs (corrections.jsonl) are read into the agent context.
Boundary markers: No specific delimiters or safety instructions are defined to separate the untrusted data from the agent's operational instructions.
Capability inventory: The skill has the capability to execute shell commands via python3 and write to local SKILL.md files.
Sanitization: There are no explicit instructions for the agent to sanitize or escape the content derived from the notes before it is used in command-line arguments.
Mitigation: The skill includes a mandatory verification step using the AskUserQuestion tool, requiring the user to review and approve the exact lines and target files before any persistence or script execution occurs.

skill-improver