The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Python module (python3 -m anibon_search) using a shell command line. The input is a JSON string passed as a single-quoted argument. This pattern creates a surface for command injection if the agent includes single quotes or other shell-special characters in the JSON values without proper escaping by the platform.
[DATA_EXFILTRATION]: The skill discloses a specific local file system path (/Users/zenithth/timestamp_workspace) belonging to the author's local environment. While not a credential, this provides information about the local system structure and username.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from a livestream archive, including markdown files and compressed transcript summaries. This data is ingested into the agent's context, creating a potential surface for indirect prompt injection if the archive content contains malicious instructions intended to influence the agent's behavior.
Ingestion points: Processes files located in timestamp_workspace/by_video_id/*.md and timestamp_workspace/transcripts/*.jsonl.gz via the search tool.
Boundary markers: No specific delimiters or safety warnings for the ingested data are defined to separate search results from instructions.
Capability inventory: The skill possesses the ability to execute shell commands via the anibon_search tool.
Sanitization: The instructions include a specific rule for 'Royal News Masking' to neutrally summarize sensitive content, which acts as a manual content filtering step.

anibon-search