The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is primarily used by AI agents to perform code reviews and refactoring tasks, which requires ingesting untrusted source code and metadata from Pull Requests (e.g., in engineering-pr-review). This creates an inherent surface for indirect prompt injection if the ingested data contains malicious instructions aimed at bypassing engineering rules or influencing agent behavior.
Ingestion points: Source code files and PR metadata processed during code review and refactoring workflows.
Boundary markers: None explicitly defined within the skill's instructional content.
Capability inventory: Includes execution of linting and testing tools (ruff, pytest, prek), file system modifications during refactoring, and network operations for status reporting.
Sanitization: No explicit sanitization or filtering logic is prescribed for the processed data.
[DYNAMIC_EXECUTION]: The rule asr-models recommends a pattern for loading custom ASR backends dynamically at runtime using import_module and getattr on a path retrieved from the CUSTOM_MODEL_PATH environment variable. This constitutes dynamic code loading from a computed path.
[EXTERNAL_DOWNLOADS]: The rules reference several external dependencies and code sources. Specifically, asr-vad describes downloading models from the Silero VAD project on GitHub (snakers4/silero-vad) via torch.hub.load, and py-ruff-format-modern references the official Ruff pre-commit repository (https://github.com/astral-sh/ruff-pre-commit). These are established and well-known sources in the development community. The skill correctly advises mocking these network-dependent calls in test environments.
[DATA_EXFILTRATION]: The rule prefect-webhook-hooks describes a mechanism for reporting workflow state changes by sending HTTP POST requests to a user-provided webhook_url. While intended for orchestration status notifications, this pattern allows for the transmission of event data to external, caller-controlled endpoints.

engineering-rules