Roadmap

Roadmap

Tooli development is organized into parallel work streams so multiple developers or agents can contribute simultaneously without file conflicts.

See PLAN.md for full issue details.

v5.0: Universal Agent Tool Interface (Complete)

The v5.0 release adds a Python API, capabilities, handoffs, and multi-framework integration:

• app.call(), app.acall() for typed in-process invocation returning TooliResult
• app.stream(), app.astream() for iterating list results item by item
• capabilities=[...] for granular permission declarations with STRICT mode enforcement
• handoffs=[...] and delegation_hint for multi-agent workflow orchestration
• Error field mapping linking errors to specific input parameters
• output_schema in envelope meta for detailed response format
• AGENTS.md generator for GitHub Copilot / OpenAI Codex compatibility
• 4 agent SDK integration examples (Claude, OpenAI, LangChain, Google ADK)
• All 18 example apps updated with v5 metadata
• 529+ tests total

v4.1: Caller-Aware Agent Runtime (Complete)

The v4.1 release adds the TOOLI_CALLER convention and caller detection module:

• 5-category heuristic detection (human, AI agent, CI/CD, container, unknown automation)
• TOOLI_CALLER environment variable convention for explicit agent identification
• detect-context builtin command on every Tooli app
• Caller metadata in envelope, telemetry, and invocation recordings
• Adaptive confirmation and help formatting for agents
• 100 new tests (359 total)

v4.0: Agent Skill Platform (Complete)

Task-oriented SKILL.md, --agent-bootstrap, PipeContract, composition inference, tooli init, metadata coverage, Claude Code integration. 73 new tests.

Completed (27 issues closed)

All of Phase 1 and significant portions of Phase 2 and Phase 3 are implemented:

• Phase 1 (15/15): Scaffolding, Tooli core, output modes, envelope, global flags, ToolError, exit codes, StdinOr, schema generation, --schema flag, command metadata, SKILL.md, contract tests, config precedence
• Phase 2 (4/13): MCP export, MCP stdio server, --response-format, --help-agent
• Phase 3 (8/10): Tool versioning, dry-run framework, eval harness, security policy, cross-platform prompt, secrets handling, telemetry pipeline

Remaining: 12 Issues in 5 Parallel Streams

  Stream A (docs)    Stream B (serving)    Stream C (arch)    Stream D (output)    Stream E (standalone)
  ┌──────────┐       ┌──────────┐         ┌──────────┐       ┌────┬────┐          ┌────┬────┬────┐
  │ #24      │       │ #19      │         │ #26      │       │#21 │#27 │          │#29 │#34 │#36 │
  │ llms.txt │       │ MCP HTTP │         │transforms│       │pag.│null│          │idem│otel│auth│
  ├──────────┤       ├──────────┤         ├──────────┤       └──┬─┴──┬─┘          └────┴────┴────┘
  │ #25      │       │ #28      │         │ #30      │          │    │               all parallel
  │ man page │       │ OpenAPI  │         │providers │       ┌──▼────▼─┐
  └──────────┘       └──────────┘         └──────────┘       │   #23  │
    parallel           parallel            sequential        │annotns.│
                                           #26 → #30        └────────┘

10 of 12 issues can start immediately (labeled parallel-ready).

Stream A: Documentation Generators

#	Issue	Status
24	llms.txt generation	Ready
25	Unix man page generation	Ready

Stream B: MCP & HTTP Serving

#	Issue	Status
19	MCP HTTP server mode	Ready
28	OpenAPI schema export & HTTP serve	Ready

Stream C: Transform & Provider Architecture

#	Issue	Status
26	Transform pipeline	Ready
30	Provider system	Blocked by #26

Stream D: Output Enhancements

#	Issue	Status
21	Pagination, filtering, truncation	Ready
27	Null-delimited list processing	Ready
23	Behavioral annotations in all output	After #21, #27

Stream E: Standalone Features

#	Issue	Status
29	--idempotency-key support	Ready
34	OpenTelemetry observability	Ready
36	Authorization framework	Ready