The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: A reference URL in the technical rationale document was flagged by automated security scanners.
Evidence: The file references/rationale.md includes a citation to https://leehanchung.github.io/blogs/2025/07/03/pydantic-is-all-you-need-for-performance-spaghetti/, which is blacklisted by URL reputation scanners. While used for technical context regarding Pydantic performance, the domain's reputation poses a risk if followed or accessed by an agent.
[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for ingesting and processing untrusted external data, creating an attack surface for indirect prompt injection.
Ingestion points: The skill defines data modeling patterns for parsing API responses, MCP tool arguments, and configuration files in SKILL.md and references/snippets/data-modeling.py (e.g., the ApiOrderRequest class).
Boundary markers: The skill encourages using Pydantic v2 with frozen=True and strict validation decorators (e.g., AwareDatetime, AfterValidator for UTC normalization) to enforce data constraints.
Capability inventory: The skill leverages capabilities for network requests (httpx), concurrent task execution (asyncio.TaskGroup), and command-line interface generation (Typer).
Sanitization: Employs Pydantic's internal validation mechanisms and recommends strict type-checking configurations (basedpyright) as primary sanitization strategies.
[DYNAMIC_EXECUTION]: The skill documents and provides examples for dynamic thread dispatching.
Evidence: In references/rationale.md and references/snippets/async-fanout.py, the skill describes the use of anyio.to_thread.run_sync and asyncio.to_thread to bridge synchronous tool calls into asynchronous event loops. This involves dynamic execution of code within thread pools.

python-conventions-taudelta