Track A: Make Neva a top-tier target for AI code generation

[emil14]

Tracking

• Parent umbrella tracker: 2026 AI Direction Tracker: Neva for AI Codegen and GenAI Apps #1056

Goal

Make Neva a top-tier language target for AI coding agents (Codex/Claude Code/Cloud Code/etc.) by reducing token waste, ambiguity, and trial-and-error during generation and review.

Problem statement

Today, AI coding agents often operate via generic text search (grep-style) and broad file reads. That is expensive and noisy. Neva can do better because the language is explicit and static, and the CLI already has an initial neva doc capability.

We should define a deliberate "AI-friendly developer tooling" layer so agents can:

• fetch only relevant symbols/components,
• understand signatures and constraints quickly,
• generate valid code with fewer retries,
• produce easier-to-review diffs.

Existing context to leverage

• CLI neva doc currently searches stdlib symbols (good base, but search-oriented): Tooling: Markdown Documentation Generation (from Neva Source Code) #973
• Visual review direction (graph/text parity): Implement Visual Editor for Source Code level (AST) Program Representation #970
• Stdlib API stabilization baseline: Stdlib usability stabilization #1033

Workstreams

1) AI-oriented CLI ergonomics

• Extend neva doc beyond grep-like output toward structured retrieval:
  • stable machine-readable output mode (for agents),
  • symbol-level lookup (component/type/const),
  • concise signature-focused responses.
• Add focused commands (or flags) for frequent agent tasks:
  • "where is this symbol defined",
  • "show public API of package",
  • "show compatible components/ports by type".

2) Tool integration surface

• Evaluate and prototype an official Neva MCP server (or equivalent tool interface) so agents can query Neva semantics directly rather than scraping raw text.
• Define minimal protocol surface for:
  • symbol lookup,
  • API docs retrieval,
  • diagnostics explanation,
  • graph-aware code navigation.

3) Deterministic generation guardrails

• Strengthen formatting/style automation and lint integration so generated code converges quickly.
• Ensure diagnostics remain actionable for iterative agent loops.
• Document "LLM coding with Neva" best practices (prompt + workflow recipes).

4) Documentation pipeline for agent consumption

• Drive Tooling: Markdown Documentation Generation (from Neva Source Code) #973 to produce deterministic Markdown/structured docs directly from .neva sources.
• Keep docs compact and reference-friendly to reduce context-window cost.

Deliverables

• Design doc with command/tool API proposals.
• At least one end-to-end prototype where an agent produces/edits Neva code using the proposed tooling surface.
• Benchmarks comparing baseline workflow vs improved tooling (tokens/time/retry count).

Non-goals

• Not building a model-specific integration for one provider only.
• Not changing core language semantics to add AI-specific syntax sugar.

Acceptance criteria

• A defined, versioned AI-tooling surface exists (CLI and/or MCP).
• Agent-driven code tasks show measurable improvement in at least one benchmarked workflow.
• Documentation and examples are sufficient for contributors to reproduce the workflow.
• Follow-up implementation issues are split and prioritized.

[emil14]

AgentOps re-trigger after webhook enable: please provide concise plan + acceptance criteria.