The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes instructions for instrumenting LLM agent frameworks (e.g., OpenAI, Anthropic, Pydantic AI) to monitor interactions and token usage. This creates an ingestion surface for untrusted data from LLM outputs.
Ingestion points: LLM API calls and agent execution methods documented in SKILL.md and references/integrations.md.
Boundary markers: The skill does not specify particular delimiters for LLM content within the monitoring context.
Capability inventory: The skill enables logging, span creation, and metric tracking across its reference files to record execution data.
Sanitization: The skill provides comprehensive examples for using logfire.ScrubbingOptions and custom callbacks in SKILL.md and references/advanced.md to redact sensitive patterns (passwords, tokens, PII) before they are logged, which is a key security mitigation.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the logfire package and its framework-specific extensions through standard package managers. These resources are established tools for Python observability.
[SAFE]: The skill promotes secure practices for secret management, recommending the use of environment variables for authentication tokens and demonstrating robust automated data scrubbing mechanisms to protect user privacy.

logfire