The Agent Skills Directory

[DATA_EXFILTRATION]: The script publish.ts reads the full content of a local file specified by the --file argument using readFileSync and sends this data to a remote host via the fetch API. This allows for the exfiltration of potentially sensitive information if the agent is subverted into reading non-public files.
[DATA_EXFILTRATION]: The skill transmits a bearer authentication token (ARTIFACT_FAIL_TOKEN) in the headers of HTTP POST requests to a destination specified by the ARTIFACT_FAIL_ENDPOINT environment variable. If the endpoint is configured to an untrusted or attacker-controlled domain, both the authentication credentials and the document data would be exposed.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a local TypeScript file (publish.ts) using the bun runtime, granting the skill the ability to perform persistent file system operations and network communication.
[DATA_EXFILTRATION]: The skill possesses an indirect prompt injection surface as it processes external file content without sanitization or boundary markers. Ingestion point: publish.ts (file content). Capability inventory: readFileSync (file read), fetch (network send). This could allow malicious instructions within processed files to influence agent behavior or facilitate data leakage.

artifact-fail