The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data, such as uploads and remote downloads, which creates a surface for indirect prompt injection vulnerabilities.
Ingestion points: The instructions and scripts process data from HTTP request bodies, remote API downloads, and child process outputs as described in SKILL.md and references/stream-safety.md.
Boundary markers: The skill explicitly mandates safety controls, including maxBytes limits, Content-Length verification, and global execution deadlines to prevent uncontrolled data ingestion.
Capability inventory: The skill utilizes file system operations (createWriteStream), network requests (fetch), and sub-process execution (spawn).
Sanitization: It implements rigorous byte-level counting, expansion ratio limits for decompression, and immediate reader cancellation for unread or non-successful response bodies to mitigate resource consumption attacks.
[COMMAND_EXECUTION]: The skill provides guidelines for spawning child processes, such as media detectors, which involves executing system commands.
Evidence: SKILL.md and evals/evals.json detail the use of spawn for media processing. The instructions focus on safety by recommending the suppression of unnecessary output streams (stderr/stdout), setting strict byte limits on required output, and enforcing process-level timeouts to prevent runaway processes.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading and uploading data to external providers as part of its core functionality.
Evidence: evals/evals.json includes scenarios for interacting with OpenAI-compatible providers and remote file sources. These network operations are standard for the skill's stated purpose and include recommendations for using duplex: 'half' and shared abort signals.
[SAFE]: The skill utilizes the @nest-boot/temporary-directory package to manage file lifecycles.
Evidence: This package is a vendor-owned resource from the skill's author (nest-boot). Its integration for managing temporary files within a defined RequestContext aligns with secure resource management and ensures automated cleanup of transient data.

nodejs-streaming