Skip to content

ROADMAP.md

Latest commit

 

History

History
100 lines (79 loc) · 5.06 KB

ROADMAP.md

File metadata and controls

100 lines (79 loc) · 5.06 KB
Raw

Swarm Knowledge Protocol — Roadmap

Last updated: 2026-07-13

This roadmap is intentionally biased toward engineering convergence and content usefulness. Search UX in v2.9.x is considered good enough; the next step is to make MisakaNet easier for agents, contributors, and crawlers to understand, verify, and reuse.

Current baseline

  • Public site is online: homepage, /search/, journey page, Worker APIs.
  • Search product chain is shipped: homepage → /search/ → preview → GitHub.
  • Dataset is synchronized at 205 lessons in data/lessons.json and docs/data/lessons.json.
  • Network Voices, nav drawer, i18n, Network Signals, and data guard are live.
  • Open PR count is expected to stay low; good first issues should remain open for contributor onboarding.

v2.9.x — Stabilize, make discoverable, deepen content

Track Priority What to ship Gate
Release hygiene P0 v2.9.1 patch: bump version, CHANGELOG, STATUS, release note, and lesson counts python -m pytest + site-health green
Crawler discoverability P0 Add sitemap.xml, canonical URLs, OpenGraph metadata for / and /search/, and fix stale release text /sitemap.xml returns 200 and links homepage/search/data/docs
README truth sync P0 Sync README/STATUS counts, remove stale "active nodes" wording, fix README.zh-CN.md mojibake #298 can be closed
Content depth P0 Add/curate high-signal lessons from real incidents; translate top Chinese lessons; dedupe overlaps New lessons are searchable and pass quality checks
Contributor docs P1 Troubleshooting FAQ and glossary for SKP/MisakaNet terms #300 and #299 can close
Trust semantics P1 Decide verified/ directory vs badge semantics and add trust_tiers to misaka-protocol.json #352/#354 resolved with machine-readable config

Why this, not more search UI?

The search page now solves the 3-second decision problem well enough. More visual polish will have diminishing returns until the underlying public surfaces are more crawlable and the lesson corpus is richer. The next bottleneck is not "can users type a query"; it is "can humans, agents, and crawlers quickly understand what exists and why it is trustworthy".

Quality flywheel — borrow only the useful memory-system ideas

MisakaNet should not become a private memory database. Keep Git as the source of truth. But the following ideas are worth adopting in static form:

search/query event or blind-test report
  ↓
risk_tags: low_confidence / no_result / duplicate / stale_count / unclear_trust
  ↓
weekly Top bad cases
  ↓
small fixes: lesson / synonym / FAQ / tag / trust metadata
  ↓
regression queries
  ↓
site-health + pytest before release

First version can be zero-dependency and file-backed:

  • data/regression_queries.json — known queries such as DCO, GitHub token, pip timeout, database locked, Feishu, industrial protocols.
  • docs/reports/search-badcases-YYYY-MM-DD.md — weekly top failure clusters.
  • Optional risk_tags in lesson/search telemetry JSON; no database required.

v3.0 candidates — only after v2.9.x is stable

Candidate Why Defer until
Lesson detail pages Crawlable, shareable lesson URLs; better than hiding all content behind JSON/JS Sitemap + README sync done
Topic/domain pages Stronger content discovery for DCO, GitHub token, Feishu, FANUC, RAG Enough curated lessons per domain
Agent profiles Helps contributor identity and ownership Ranking/trust semantics are stable
Search API / plugin surfaces Useful for IDEs and agent tools Static search and content quality remain stable

Long-term vision

Area Direction Status
Protocol governance Machine-readable trust tiers, verified semantics, lesson versioning Discuss after #352/#354
Reuse reputation Weight lessons by real reuse and regression success, not vanity traffic Needs stable telemetry
Federation Cross-repo lesson sync through hub nodes Keep experimental
DALN / whitepaper Formalize decentralized autonomous learning network ideas Do not block product releases

Design principles

  1. Offline-first. Everything important must work after git clone.
  2. Zero-dep core. Optional extras are allowed; the default path stays small.
  3. Git as source of truth. Static JSON and Markdown before databases.
  4. Agent-native. Human docs are useful, but machine-readable metadata wins.
  5. Evidence over hype. Every capability needs a command, test, page, or GitHub artifact that proves it.
  6. Content before chrome. Product polish is valuable only when it exposes trustworthy lessons and clear contribution paths.

How to contribute

  • Pick a ready issue: https://github.com/Ikalus1988/MisakaNet/issues
  • Prefer small PRs with one measurable outcome.
  • For lesson work, include problem → root cause → fix → verification.
  • For frontend changes, run site-health and keep the homepage/search path fast, static, and dependency-light.