A local-first, python-standard-library-centric AI autonomous agent operations plane prioritizing secure offline execution without sacrificing automation.

• Local First: Safe offline command operation, deterministic text evaluation metrics.
• Python Standard Library: No mandatory third party web framework layers for internal execution logic.
• Enterprise Safety Guards: Action approvers requiring dry-run manual reviews, redaction filters for secrets, and dynamic execution profiles.
• Hermes-Compatible Skills: Marketplace manifests can model Hermes Agent SKILL.md workflows while preserving ZAI Coder policy gates.

• Fully transparent task queuing.
• Offline RAG chunking and lexical database indexing.
• Standalone deploy planner that never auto-mutates hosts.
• Embedded continuous observability metrics.
• Policy-gated Agent Marketplace with skill manifests, compatibility checks, install/enable approvals, and audit logs.
• No git add . allowed.

make install-dry-run
./zai-coder self heal --check
make package APPLY=1

ZAI Coder includes an Agent Marketplace layer for reusable local skills. Skills are tracked through SkillManifest records, can be searched by category/tags/source, and are designed to align with the Hermes Agent SKILL.md model:

make skill-catalog
make marketplace-search QUERY=hermes
make skill-install-policy SKILL_ID=hermes-skill-author AGENT_TYPE=builder

See:

• docs/marketplace/AGENT_MARKETPLACE_AND_SKILLS_GUIDE.md
• docs/marketplace/SKILL_MANIFEST.md
• docs/marketplace/HERMES_SKILLS_ALIGNMENT.md
• skills/hermes-skill-author/SKILL.md

Refer to the docs/ folder for explicit CLI usage maps, system diagrams, and regression pipeline outlines.