Roadmap

Maturity Levels

Label	Meaning
Stable	API is unlikely to change. Covered by tests and used in production workflows.
Alpha	Functional and tested, but the API may change before v1.0.
Experimental	Works end-to-end but may have rough edges. Feedback welcome.
Planned	Designed but not yet implemented.

Implemented in Repo

All @aex-lang/* packages are published on npm (v0.0.4). APIs are not yet stable — expect breaking changes before v1.0.

1. Runtime (Alpha)
   • [x] Built-in checks (patch touches only, diff linting)
   • [x] Tool registry (file.write, tests.run, git.diff, git.apply)
   • [x] Command injection and path traversal prevention
   • [x] JSON IR compilation
2. Adapters (Alpha)
   • [x] OpenAI Agents SDK (@aex-lang/openai-agents)
   • [x] MCP gateway (@aex-lang/mcp-gateway)
   • [x] LangGraph compiler (@aex-lang/langgraph)
3. Developer experience (Alpha)
   • [x] aex fmt auto-formatter
   • [x] CLI diagnostics with error codes
   • [x] VS Code extension (syntax highlighting, snippets)
   • [x] Interactive playground on docs site
4. Security (Experimental)
   • [x] Threat-model reference implementation
   • [x] Signed contracts (aex sign / aex verify)
   • [x] Timing-safe HMAC verification
5. Model handlers (Experimental)
   • [x] Built-in OpenAI handler (AEX_MODEL=openai)
   • [x] Built-in Anthropic handler (AEX_MODEL=anthropic)
   • [x] Custom handler support (--model-handler ./path.ts)
   • [x] Budget enforcement at runtime (cap do/make invocations)
6. Control flow (Alpha)
   • [x] if conditional branching with indentation-based blocks
   • [x] for loop iteration over lists
   • [x] Nested control flow (if inside for, etc.)
7. Remote tool registries (Experimental)
   • [x] aex run --registry <url> fetches tool definitions from HTTP endpoints
   • [x] Remote tools execute via POST with JSON args/response
8. Policy inheritance & composition (Alpha)
   • [x] extends field in policies (file path or inline object)
   • [x] composePolicies() merges allow/deny/confirmation/budget
   • [x] Budget takes the minimum across composed policies
9. Structured logging & OpenTelemetry (Experimental)
   • [x] createStructuredLogger() with timestamps, traceId, spanId
   • [x] --log-json flag for JSON event output
   • [x] --otlp-endpoint flag exports traces in OTLP format
   • [x] exportToOTLP() API for programmatic export
10. CI & publishing (Alpha)
    • [x] setup-aex GitHub Action (action/action.yml)
    • [x] npm publish config (publishConfig, exports, repository on all packages)
    • [x] scripts/prepublish.sh rewrites file: deps to versioned refs
11. Policy files & merge semantics (Alpha)
    • [x] policy workspace v0 keyword for ambient security boundaries
    • [x] aex init --policy scaffolds .aex/policy.aex
    • [x] Merge semantics: allow = intersection, deny = union, confirm = union, budget = min
    • [x] aex effective previews merged permissions before running
    • [x] Parser-level validation: policy files reject need, do, make, check, return
12. MCP proxy (Alpha)
    • [x] aex proxy -- <cmd> gates MCP tool calls against policy
    • [x] Auto-discovers .aex/policy.aex
    • [x] Budget enforcement, confirmation gates, allow/deny filtering
    • [x] Structured JSON audit logging to stderr
    • [x] tools/list response filtering
13. Claude Code hook enforcement (Alpha)
    • [x] aex gate PreToolUse hook gates built-in tools (Read, Write, Bash, etc.)
    • [x] Tool name mapping: Claude Code PascalCase → AEX dotted capabilities
    • [x] Budget state persistence across hook invocations
    • [x] allow keyword for policy files, AEX120/AEX121 diagnostics
    • [x] task keyword as alternative to agent

• [x] Fail-closed by default (denies all when no policy found)

14. Draft → Review → Run workflow (Alpha)
    • [x] aex draft generates task contracts from natural language prompts
    • [x] aex review shows human-readable contract summary with effective permissions
    • [x] aex review --run prompts for approval then executes through runtime
    • [x] aex classify classifies prompts as exploratory/contract_recommended/contract_required
    • [x] .aex/runs/ directory for generated one-off contracts
    • [x] Audit log output (.audit.jsonl) for executed contracts
    • [x] aex run auto-discovers .aex/policy.aex and supports .aex policy files
    • [x] LLM-powered contract generation with validation loop and retry

Up Next

• Stable API guarantees (v1.0 milestone)
• Independent security audit

Not Yet Done

• No stable API guarantees
• No independent security audit

Contributions are welcome — open an issue if you'd like to tackle an item or propose a new milestone.