Pack version: v1.0.0 · Last updated: 2026-05-15
Languages: English | 中文
This is an independent AI context resource pack for microsoft/agent-framework: host instructions, prompt preview, evals, pitfalls, and recovery rules you can load into Claude Code, Codex, Cursor, Aider, and other AI coding agents.
- Copy
AGENTS.mdinto your AI coding host (orCLAUDE.mdfor Claude Code). - Run the first prompt in
01_PROMPT_PREVIEW.md. - Verify behavior with
06_EVALS/smoke_check.md. - If verification fails, recover with
03_PITFALL_LOG.md.
Quick links: Start · Prompt · Evals · Pitfalls · Manual
Use this pack when you want an AI coding agent to understand the upstream project context, setup boundaries, common pitfalls, and verification steps before it edits files, installs dependencies, or claims the tool is ready.
AGENTS.md/CLAUDE.mdfor AI coding agent hosts.- Prompt preview in
01_PROMPT_PREVIEW.md. - Acceptance checks in
06_EVALS/. - Pitfall log and recovery steps in
03_PITFALL_LOG.md. - Boundary and risk card in
04_BOUNDARY_RISK_CARD.md. - Source attribution in
SOURCE_MAP.md.
See AGENTS.md for host instructions. Claude Code reads .md files in the project root as agent instructions.
Top 3 most actionable:
- .NET SDK version conflict — causes MissingMethodException or FileNotFoundException at runtime; verify SDK version matches agent-framework release requirements.
- Model configuration ignored — agent uses wrong model or ignores API key; set
AGENT_MODEL_*env vars before runtime init. - Multi-agent initialization failure — agents hang or CustomStatus overflow; use separate runtime instances per agent role.
Real failure modes and recovery paths are documented in 03_PITFALL_LOG.md.
Pack assembled by Doramagic. Upstream: https://github.com/microsoft/agent-framework (MIT license).
- Upstream/source: https://github.com/microsoft/agent-framework
- License: MIT
- Pack contents: prompts, host instructions, checks, guardrails, and validation notes
- Relationship: independent pack; not affiliated with or endorsed by microsoft/agent-framework unless explicitly stated
If this pack helps your agent work from evidence, star the repo or open an issue for bugs, usage questions, or new pitfall reports.