Skip to content
Oak Dev-inter edited this page May 2, 2026 · 8 revisions

Kasidit Wiki

Mindful AI coding framework. Discipline over cleverness. Works on any model tier.

Kasidit is a Claude Code plugin that makes AI coding grounded instead of hallucinated. It scaffolds the discipline a senior engineer carries in their head β€” mission narrowing, design-before-code, runtime-as-judge, confidence labels, checklists, and tier-aware behavior β€” into an operational framework any model can follow.

Start here

  • Installation β€” install the plugin, run the v0.10 backend installer, init a project.
  • Getting Started β€” your first mission in under 5 minutes.
  • Commands β€” full reference for every command (Mode + missions + Gravity).
  • Kasi-Mode β€” /kasi off|router|lite|full|ultra (v0.10).
  • Backend-Hooks β€” runtime hooks (v0.10).
  • Gravity Pattern β€” two-tier knowledge system (Centerlite + Dcenterlite).
  • Model Tiers β€” what changes on Opus vs Sonnet vs Haiku.
  • FAQ β€” common questions.

Community

Version history

Deep per-release notes, in descending order:

  • v0.13.0 β€” thClaws Runtime Support (Consolidated). Supersedes v0.12.0/v0.12.1. Single clean install path. Kasidit now runs on both Claude Code and thClaws. Same SKILL.md, 21 commands, 8 agents, 15 checklists. ~85% feature parity on thClaws.
  • v0.12.0 β€” thClaws Runtime Support. Kasidit now runs on both Claude Code and thClaws. New install-thclaws.sh, mirrored .thclaws-plugin/ manifests, hook event mapping (4/5 hooks adapted), docs/thclaws-setup.md. ~85% feature parity on thClaws.
  • v0.11.0 β€” Backend Mission Router + Structure Bridge + Runbook Capture. 6 new commands (/kasi-backend, /kasi-graph, /kasi-struc, /kasi-devopt, /kasi-acknowledge, /kasi-knowledge-list), 3 new backend checklists (Laravel Β· Node Β· API design), 4 helper scripts (function call graph + state cache writers). Hooks renamed kasidit-* β†’ kasi-* (history preserved; emit-token protocol kept).
  • v0.10.0 β€” Honesty Cleanup. SKILL-full split reverted (Mode-gated single file), audit-specialist consolidates code-reviewer/security-auditor/perf-profiler, runtime backend hooks (route/verify/record/update/drift), install.sh canonical installer, 12 default checklists, incremental backend save (route-memory.jsonl).
  • v0.9.2 β€” Gravity Pattern. Two-tier knowledge (Centerlite + Dcenterlite), global prompt log, /kasi-init, promote/pull/sync commands, wiki sync.
  • v0.9.1 β€” Master Orchestrator Rule. 7 specialist agents; master delegates all strong work.
  • v0.9.0 β€” Claude Design Integration. Route visual missions to Claude Design; stop hand-coding mockups.
  • v0.8.0 β€” Tier Cascade orchestration + local embedding knowledge layer.
  • v0.7.4 β€” SWE-bench validation + Rules 2.3–2.6 (no fake metrics, numbered options, native language, mandatory git-log on bug fix).
  • v0.3.0 β€” Model tier adaptation, confidence labels, CHECKLISTS/, PATTERNS.md, multi-agent orchestration, verifier pass.
  • v0.2.1 β€” Documentation retrieval protocol (trust hierarchy, version-matched docs, knowledge caching).
  • v0.2.0 β€” UI Override Mode + cache protocol + domain detection.
  • v0.1.0 β€” Core principles, mission counter, ΰΈͺารบัญ (index) system.

See Version History for a side-by-side comparison table.

Why Kasidit exists

AI coding fails not because models are too small, but because they lack a grounded base.

The fix is not a bigger model β€” it is discipline externalized into files, checklists, and commands. Kasidit provides that scaffold.

License + source

Kasidit

Core

Version History

Concepts

Commands

Agents

Deprecated v0.10 (stubs β†’ audit-specialist --focus=..., removed in v0.11)

Clone this wiki locally