The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides a direct path for extracting data from the local filesystem to a remote Telegram chat via the alma CLI tool.
Evidence: The skill's instructions tell the agent it must deliver "ANY file the user asks to see" and mentions "sharing any file path".
Evidence: The bash commands alma send file, alma send photo, etc., transmit the file contents to the destination specified by the ALMA_CHAT_ID environment variable.
[COMMAND_EXECUTION]: The skill is configured to use the Bash tool to execute system commands.
Evidence: The allowed-tools section explicitly includes Bash, and the core functionality relies on executing the alma command-line interface.
[DATA_EXFILTRATION]: The skill has a high vulnerability to data exposure via user manipulation (Indirect Prompt Injection surface).
Ingestion points: File paths are ingested directly from user prompts or generated by the agent based on user requests (e.g., SKILL.md mentions triggers like "sharing any file path").
Boundary markers: None. There are no instructions or system constraints provided within the skill to prevent the agent from accessing sensitive locations like ~/.ssh/, .env files, or /etc/.
Capability inventory: The skill combines filesystem read access with network transmission capabilities via the alma tool.
Sanitization: There is no evidence of path validation or sanitization logic to ensure that only intended or safe files are transmitted.

send-file