Skip to content

roadmap-rfc-issue.md

Latest commit

 

History

History
113 lines (86 loc) · 3.25 KB

roadmap-rfc-issue.md

File metadata and controls

113 lines (86 loc) · 3.25 KB
Raw

[Roadmap RFC] Evaluate MisakaNet's next 90-day direction

Objective

MisakaNet is evolving from a zero-dependency lesson library into an agent failure-memory network.

We need a concrete 90-day strategy memo that evaluates MisakaNet's next development direction.

Current candidate visions:

1. Experience reuse substrate

  • Git-backed reusable lessons
  • OKF-compatible lesson format
  • SAG-Lite search
  • Zero-dependency local search

2. Lightweight MCP gateway

  • Thin MCP server for agent/tool access
  • Claude Code / Cursor / Continue.dev / other MCP clients
  • MisakaNet as a tool layer, not a heavy platform

3. Agent capability testing platform

  • Use real failure lessons to test whether agents can recover
  • Measure search, reuse, fix quality, and contribution quality
  • Avoid becoming a generic LLM benchmark

4. Other directions

  • Contributor reputation
  • Helpful/reuse feedback loop
  • Domain-specific cookbooks
  • Field reports
  • Tool integration ecosystem

Deliverable

Submit a markdown strategy memo under:

docs/roadmap/90-day-strategy-YYYY-MM.md

The memo should include:

  • One-sentence positioning of MisakaNet
  • Evaluation of at least 3 candidate directions
  • Evidence from current repo state:
    • lessons count
    • stars/forks
    • open issues
    • existing integrations
    • current docs/content structure
  • Decision matrix:
    • impact
    • effort
    • risk
    • fit with zero-dependency/local-first principles
  • What to double down on
  • What not to do
  • Suggested 30/60/90-day roadmap
  • Suggested next 5 issues to open
  • Release criteria for the next version

Suggested framing

MisakaNet should not become "another RAG app".

Preferred framing to evaluate:

  • OKF-compatible lessons as the knowledge format
  • SAG-Lite as the search layer
  • MCP as the agent access layer
  • Real failure recovery as the evaluation loop

Out of Scope

  • Full vector database migration
  • Generic LLM leaderboard
  • Heavy SaaS platform
  • Replacing the zero-dependency CLI path
  • Vague marketing-only proposals

Acceptance Criteria

  • Evaluates at least 3 strategic directions
  • Clearly separates core vision from optional integrations
  • Keeps zero-dependency local search as a protected baseline
  • Explains whether MCP should be core, adapter, or optional layer
  • Explains whether MisakaNet should become an agent benchmark platform
  • Includes a 30/60/90-day roadmap
  • Includes at least 5 concrete follow-up issues
  • Includes a "do not build yet" section
  • Uses current repo evidence, not generic AI hype
  • Written in clear English, with optional Chinese summary

Priority

P1 / roadmap / bounty / ready

Not P0. P0 remains:

  1. OKF/SAG-Lite integration
  2. MCP thin server verification
  3. Helpful/reuse feedback loop closure
  4. #277 real lesson merge

Competition Rules

  • 48h soft deadline. Auto-closes if no update for 48h.
  • Claim window: 4 hours. Once claimed, 4 hours exclusive lock.
  • First pass-CI PR wins. If CI-green and passes review, it gets merged.

⚠️ Maintainer Note

  • No bounty. Merge is the only reward.
  • This is a zero-bounty, high-merit autonomous competition.
  • Missing Node ID in PR frontmatter → auto closed.
  • PRs that fail any mandatory AC will be closed without review.