Skip to content

ROADMAP.en.md

Latest commit

 

History

History
81 lines (55 loc) · 5.62 KB

ROADMAP.en.md

File metadata and controls

81 lines (55 loc) · 5.62 KB
Raw

Roadmap

This document tracks the medium-term and long-term direction of boss-agent-cli. Issues and PRs are welcome for any of the areas below.

Released

  • ✅ v1.8.x (2026-04-20, rapid patch series): strict mypy whitelist expanded from 3 to 61 modules (81% coverage), Python embedding API, ai interview-prep / chat-coach, Markdown digest output, Cursor/Windsurf integration, four additional AI provider entry points, and the English contribution guide
  • ✅ v1.8.0 (2026-04-19): AI communication and interview expansion (ai interview-prep / ai chat-coach) plus protocol server growth to 43 tools
  • ✅ v1.7.0 (2026-04-17): draft chat replies, application funnel analytics, and protocol server growth to 41 tools

Full release history lives in CHANGELOG.md.

Near term (current mainline)

Data visualization

  • boss stats --format html outputs an interactive funnel report (v1.7.1)
  • boss digest --format md can be pasted directly into email or Feishu workflows (v1.8.1)
  • codecov badge integrated into the README (v1.7.1)

Agent integration

  • MCP server supports HTTP streaming, SSE, and stdio transports (2026-04-27, PR #160)
  • host-specific integration examples for Codex, Cursor, and Windsurf (v1.8.1, docs/integrations/ now fully covered)
  • OpenAI Functions export via boss schema --format openai-tools (v1.7.1)

AI capabilities

  • boss ai chat-coach - communication guidance derived from chat history (v1.8.0)
  • boss ai interview-prep - mock interview generation based on the JD (v1.8.0)
  • support for current Claude 4.7 / GPT-5 generation models (v1.8.2, providers extended to openrouter, qwen, zhipu, and siliconflow)

Mid term (v2.0)

Architecture evolution

  • full mypy strict-mode rollout - 100% complete (66/66 business modules now enforce disallow_untyped_defs + disallow_any_generics + warn_return_any, v1.9.1)
  • exported type signatures in stubs/ for downstream IDE consumers (v1.8.6, including py.typed, canonical __all__, and 16 contract tests)
  • evaluate a Bridge protocol move from HTTP/WS to gRPC - research completed (Issue #96, docs/research/bridge-grpc.md), with the current conclusion set to do not migrate yet because localhost single-user scenarios do not gain meaningful performance, MV3 extension compatibility risk stays high, and dependency size would grow by about 8 MB. Five re-evaluation triggers are already documented.

Ecosystem expansion

  • Web UI (React + Tailwind) for non-agent users
  • browser extension with deeper integration into the native BOSS Zhipin pages
  • multi-platform support for Lagou / Zhilian / Liepin adapters - API research is fully complete (Issue #90 closed, docs/research/platforms/). Conclusion: Zhilian candidate-side is implemented (read-only search/detail/recommend/user_info + write greet/apply, see the Week 2-3 sub-items below); Lagou and Liepin are not pursued after evaluation; 51job remains in the research backlog.
    • Week 1a: Platform ABC skeleton + BossPlatform adapter (#129, zero behavior change)
    • Week 1b: global --platform CLI option + get_platform_instance helper + schema exposure of current_platform
    • Week 1c: command-layer migration to the Platform interface (20 commands: greet, apply, batch-greet, interviews, detail, show, me, recommend, chat, chatmsg, mark, exchange, pipeline, digest, search, export, chat_summary, history, status, watch)
    • Week 1d: ZhilianPlatform stub registered in the platform registry (abstraction self-proof with full envelope adaptation; P0/P1/P2 still raise NotImplementedError)
    • Week 2: Zhilian read-only implementation (search, detail, recommend, user_info)
    • Week 3: Zhilian write operations (greet, apply) + docs + MCP adaptation
    • Week 4: product decision on recruiter-side support and whether to onboard it at all (candidate-side support is the only runtime path today)
    • 51job: keep it in the research backlog until the candidate-side read-only entry points and redacted fixtures are clear enough for a runtime adapter (docs/research/platforms/51job.md)

Community building

  • richer Chinese and English demo assets and launch materials (the repo already ships demo/demo-zh.gif / demo/demo-en.gif plus the matching demo/demo-zh.tape / demo/demo-en.tape terminal demos)
  • follow up on the review outcome of awesome-agents PR #423
  • decide later whether to pursue awesome-claude-code
  • English contribution guide (CONTRIBUTING.en.md, v1.8.3)

Long-term vision

Make AI agents feel like real job-search copilots, not just tool-call wrappers:

  • agents should autonomously complete the full loop from search to screening to greeting to follow-up to interview prep
  • users should only need to describe a target, such as "find remote Python roles above 30K", and let the agent execute the workflow
  • all data should remain local-first, with privacy and compliance treated as primary constraints

How to contribute

  1. Pick up an item labeled good first issue or help wanted
  2. Open an issue when you want to discuss a direction or design before implementation
  3. Send a PR directly for bugs or documentation fixes
  4. Non-code help is also useful: test reports, usage feedback, and translations

See CONTRIBUTING.en.md.

The roadmap is a living document and should be updated alongside each minor release.