The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes patterns for processing external data (prompts and message arrays) through LLM interfaces while possessing capabilities to interact with the file system and network.
Ingestion points: The enqueueChat function (Step 4) and process_batch function (Step 5) accept arbitrary prompt/message data.
Boundary markers: The code snippets do not implement delimiters or specific instructions to ignore embedded commands within the user-provided prompts.
Capability inventory: The skill uses allowed-tools including Write (Step 2 uses writeFileSync to create batch files) and the environment allows network requests (Step 4 uses fetch).
Sanitization: There is no evidence of input validation or prompt sanitization for the external data ingested.
[DATA_EXFILTRATION]: Step 4 demonstrates a background worker pattern that transmits LLM results to an external endpoint.
Evidence: The worker in Step 4 uses fetch(job.data.callbackUrl, ...) to send job results to a URL provided in the job metadata. While a standard pattern for webhooks, this allows sending data to arbitrary non-whitelisted domains.

mistral-webhooks-events