feat(cli): agent-native inspection surfaces

[escapeboy]

Summary

Five read-only, --json-capable CLI commands so AI agents can inspect Boruna projects without reading source. Motivated by a competitive review of vercel-labs/zero — see claudedocs/research_zerolang_vs_boruna_2026-05-17.md.

Command	Purpose
boruna lang codes [--json]	Registry of stable diagnostic codes (E001–E009) — name, summary, category
boruna doctor [--json]	Environment/toolchain health: version, compiled features, Rust toolchain, data-dir writability, project layout
boruna workflow graph <dir> [--json]	DAG facts: nodes (kind, capabilities, deps), edges, topological order, roots, leaves
boruna size <file.ax> [--json]	Bytecode artifact cost: per-function opcode counts, totals, serialized .axbc byte size
boruna skills list / get <name>	Embedded, agent-curated docs (ax-language, cli, workflows, diagnostics) — usable with no repo checkout

All five are read-only. No persisted-schema changes, no determinism/replay impact, no new MCP tools.

Notes

• The research's idea docs: update CLAUDE.md with test count, CI/CD, and security invariants #1 ("stable diagnostic codes") was already implemented (E001–E009). Per the agreed scope it was replaced with the machine-readable registry (tooling/src/diagnostics/registry.rs) — the agent-facing piece that was missing. A drift test parses diagnostics/mod.rs so a code added without a registry entry fails the build (convention §33).
• boruna lang codes lives under lang (already "diagnostics, repair") rather than a new top-level command, to avoid surface bloat.

Test plan

• ✅ 12 new integration tests in crates/llmvm-cli/tests/cli_agent_native.rs
• ✅ 3 registry drift/uniqueness/population tests + 3 skills tests
• ✅ cargo clippy --workspace --all-targets -D warnings clean
• ✅ cargo fmt --all --check clean
• ✅ cargo test --workspace — all 46 test targets pass except cli_watch::watch_reruns_on_file_change, verified pre-existing: it fails identically on clean master (filesystem-watch timing test, unrelated to this branch).

Design / architecture / test plan: docs/design-agent-native-cli.md, docs/architecture-agent-native-cli.md, docs/test-plan-agent-native-cli.md.

🤖 Generated with Claude Code