The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/ingest.ts script utilizes readFileSync to access files at paths provided via command-line arguments and fetch to transmit the base64-encoded content to a remote URL defined by the CONVEX_URL environment variable. This capability enables the transmission of arbitrary local file contents to a remote server.
[COMMAND_EXECUTION]: The skill requires the execution of a Bun script (scripts/ingest.ts) to handle file processing and network requests based on agent-supplied data.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external data.
Ingestion points: Data enters the system via filePath, imagePath, and external url parameters as documented in SKILL.md.
Boundary markers: There are no instructions to use delimiters or ignore embedded commands when processing content from these external sources.
Capability inventory: The agent has access to scripts/ingest.ts, which can read local files and perform network POST operations.
Sanitization: No evidence of validation or filtering is present for the prompt text extracted from external images or URLs before it is saved to the database.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install and update the skill from a GitHub repository (https://github.com/Michailbul/laniameda-gallery). This is treated as a standard vendor resource.
[CREDENTIALS_UNSAFE]: The file SKILL.md contains a hardcoded user identifier (278674008) labeled as a required Telegram user ID for the knowledge base owner.

laniameda-gallery-ingest