deepagents==0.7.0a6

Warning

This is a pre-release version. Install with: pip install deepagents==0.7.0a6

Released from alpha/deepagents-0-7-0a6 at commit 55983b3

This alpha adds a dedicated NVIDIA Nemotron 3 Ultra harness profile and adds fixes for skill metadata parsing, truncation guidance, context-aware middleware typing, and composite filesystem routing.

Features

• Add a built-in NVIDIA Nemotron 3 Ultra harness profile across NVIDIA/ChatNVIDIA, Baseten, Fireworks, OpenRouter, Nebius, and Together. The profile adds model-specific prompt guidance, provider compatibility shims, text-tool-call repair, reasoning-tag cleanup, filesystem retry handling, rate-limit retries, loop-control nudges, and final-answer guards for agentic workloads without changing other models on those providers (#4192) (d5a60ec).

Bug Fixes

• Accept YAML list values as well as comma-separated strings for skill allowed-tools frontmatter, trimming valid list entries and continuing to warn only for unsupported scalar values (#4140) (d62534c).
• Make SkillsMiddleware truncation warnings actionable by reporting the field name, file path, actual length, configured limit, model-selection impact, and the specific SKILL.md frontmatter field to shorten (#4141) (2f5f5b8).
• Preserve ContextT through the create_deep_agent(..., middleware=[...]) type annotation, so type checkers accept context-aware middleware when a matching context_schema is passed (#4055) (7be76c7).
• Correct CompositeBackend.glob and CompositeBackend.aglob routing so explicit default-backend paths such as /tools do not also return files from routed backends such as /memories (#4531) (cbdb0a7).

---

Thanks to our community contributors: Kavish Kartha (@kavishkartha05) (Twitter, LinkedIn), nv-kasikritc

Internal maintainers: Nishitha M (@imnishitha), Srimanth Tangedipalli (@srimanthtangedipalli-eng), John Kennedy (@jkennedyvz), ccurme (@ccurme), Mason Daugherty (@mdrxy)

Released by: Nishitha M (@imnishitha)