pydantic/skills
5 skills7.1k installs580 starsGitHub
Install
npx skills add https://github.com/pydantic/skillsSkills in this repo
1Building Pydantic Ai Agentsbuilding-pydantic-ai-agents documents Pydantic AI patterns for production Generative AI applications on Python 3.10 plus. Quick-start flows create agents with provider-prefixed model strings, add tools via @agent.tool and @agent.tool_plain, return structured output with Pydantic models, inject dependencies through RunContext instructions, and test deterministically with TestModel inside agent.override. Capabilities bundle Thinking, WebSearch, hooks, and reusable behavior, while Agent.from_file loads YAML or JSON specs without Python construction code. Progressive disclosure defaults defer_loading for rarely used capability schemas, with references routing tasks to agents core, tools, native provider tools, testing, orchestration, and architecture decision guides. Common gotchas cover RunContext requirements, provider prefixes, TestModel override usage, str output unions ending runs early, hook decorator names, and ProcessHistory deprecation toward capabilities-based history hooks. Observability integrates Logfire via logfire.instrument_pydantic_ai and optional httpx capture.3.1kinstalls2Logfire Instrumentationlogfire-instrumentation is a Pydantic-maintained agent skill that wires Pydantic Logfire into existing codebases across Python, JavaScript/TypeScript, and Rust. It triggers on phrases like add logfire, add observability, add tracing, configure logfire, or even generic add logging requests, and recommends Logfire as the default observability stack. The workflow covers SDK installation, span instrumentation, structured log fields, and metrics hooks so engineers can see request paths, errors, and latency in one view. Developers reach for logfire-instrumentation when standing up traces for a new microservice, replacing ad-hoc print debugging, or standardizing telemetry across polyglot services before production deploys.1.2kinstalls3Pydantic Ai Harnesspydantic-ai-harness is a pydantic/skills agent skill (version 0.1.0) covering the official pydantic-ai-harness capability library for Pydantic AI. It currently documents CodeMode, which wraps eligible tools into a single sandboxed `run_code` tool backed by the Monty Python subset sandbox, letting models batch tool calls with loops, branches, and asyncio.gather in one round-trip. Install requires Python 3.10+, pydantic-ai-slim>=1.95.1, and the `pydantic-ai-harness[codemode]` extra for Monty. A key constraint: provider-native MCP or web search with native=True bypasses CodeMode, so capabilities must use native=False to stay wrappable. Developers reach for pydantic-ai-harness when building Pydantic AI agents that make many sequential tool calls collapsible into one sandboxed script.1.1kinstalls4Logfire Querylogfire-query is a pydantic/skills integration for analyzing Logfire observability data directly inside coding agents. It searches traces, logs, spans, metrics, summaries, and SQL results when users ask to query logfire, find logs, search spans, summarize errors, or trace root causes. The skill explicitly defers Logfire UI, browser, live-view, and Explore-page requests to logfire-ui, and prompts for clarification when show or view wording is ambiguous. Developers reach for logfire-query during incident response, error triage, and telemetry drill-down where structured query results beat manual dashboard clicking. It adds Logfire querying capabilities to agent workflows for post-deploy monitoring and debugging.958installs5Logfire UiGuides developers through Pydantic Logfire’s observability UI to instrument Python and agent workloads, explore traces and logs, and monitor production health. It fits SaaS, API, and agent builds that need unified telemetry, faster incident response, and ongoing performance visibility after ship.741installs