The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (images, documents, audio, and video files). This creates a vulnerability surface where malicious instructions embedded within these files could influence the agent's output or subsequent actions.
Ingestion Points: User-provided files via the --input flag in SKILL.md and references/arkcli-understand.md.
Boundary Markers: None present for file content.
Capability Inventory: CLI command execution (arkcli +understand), network communication via API, and local file access.
Sanitization: No explicit sanitization or validation of the content of uploaded files is mentioned.
[DATA_EXFILTRATION]: The skill facilitates the transmission of local file data to external API endpoints for processing. While this is the core functionality, it introduces a risk of data exposure.
Evidence: Use of the --input @<path> flag to upload local files as described in SKILL.md and references/arkcli-understand.md.
Risk: Sensitive local files could be uploaded and stored on remote servers if the agent is directed to process inappropriate file paths.
[COMMAND_EXECUTION]: The skill relies on the execution of the arkcli binary and its subcommands to perform tasks.
Evidence: Metadata in SKILL.md specifies a requirement for arkcli.

arkcli-understand