The Agent Skills Directory

[SAFE]: A comprehensive security audit of the skill was performed, covering all 11 threat categories. No evidence of prompt injection, data exfiltration, obfuscation, or malicious persistence was found. The skill follows standard practices for managing local services and handles configuration through local files. All network operations are restricted to localhost (127.0.0.1).
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where it ingests data from local configuration files and tool outputs.
Ingestion points: The skill instructions read configuration parameters from .agent/learning/vector_profiles.json and ingest tool output from curl heartbeat checks at 127.0.0.1:8110.
Boundary markers: No delimiters or boundary markers are defined to isolate the external configuration data from the agent's instructions.
Capability inventory: The skill uses the Bash tool for command execution (starting the server and heartbeats) and the Write tool for file modifications.
Sanitization: There is no evidence of sanitization or validation of the host, port, or path parameters retrieved from the profile configuration before they are interpolated into shell commands.

vector-db-launch