List view
Package, document, and publish Kata v1 to npm as @withkata/core. ## Scope - Documentation audit — Mintlify site accuracy, broken links, current content - Skill file accuracy — verify .kata/skill/ matches actual CLI behavior - System guide update — reflect dogfooding-era and San-Ma changes - npm README — installation, quick start, what it does - LICENSE file (MIT) - CONTRIBUTING.md - Changelog generation (git-cliff) - package.json audit — verify npm pack ships the right files - npx installation experience testing - Version bump, GitHub release, npm publish - Close or triage all remaining open issues ## Exit Criteria `npm install -g @withkata/core && kata init` works for a new user. Docs are accurate. Package is published. Open issues are triaged. ## Depends On - Dogfooding & Stabilization (鍛錬) — system is stable and validated - San-Ma (三間) — three-space architecture is the v1 storage model - Bunkai Evolution (知の進化) — intelligence layer enhancements included in v1
No due date•0/1 issues closedContinued dogfooding of Kata on real projects. Fix bugs, polish UX, harden CLI, and accumulate enough real keiko cycles to validate the system before structural changes. ## Scope - All bugs and UX issues discovered during dogfooding - CLI command audit — systematic testing of every command, flag, and condition state - Agent workflow validation — kataka lifecycle, attribution, onboarding - Human experience polish — TUI, interactive flows, error recovery - Intelligence loop validation — learning → behavior improvement arc across cycles ## How It Works This is a living milestone. As we dogfood, new issues get filed and added here. The milestone is "done" when: 1. All critical/blocking bugs are resolved 2. Every CLI command has been exercised and verified 3. The agent workflow (cycle → execute → observe → cooldown) runs smoothly 4. Enough keiko cycles (target: 5-10 more) have been completed to validate the learning loop ## Dogfooding Projects (from dogfooding-roadmap.md) - P1 (Critical Fixes): Mostly done — residual issues remain - P2 (CLI Audit): Systematic command coverage testing - P3 (Agent Workflow): Kataka registration, attribution, onboarding - P4 (Human Experience): TUI, interactive flows, error states - P5 (Intelligence): Multi-cycle learning validation - P6 (Diary & Dojo): Three-part diary, collaborative cooldown ## Precedes - San-Ma (三間) — three-space architecture requires stable foundation - Bunkai Evolution (知の進化) — intelligence enhancements build on stabilized system
No due date•32/58 issues closedPost-dogfooding intelligence layer enhancements in TypeScript, before Rust port. Informed by Doodlestein ecosystem research (CASS Memory, Meta Skill, TOON, UBS) conducted in Tankyu. Adopts patterns, not dependencies. ## Epics 1. **TOON Output Format** — Minimal token-optimized encoder, `--format toon` across agent-facing CLI outputs 2. **Knowledge System Enhancements** — Anti-pattern inversion, asymmetric weighting, inline feedback, evidence-gated promotion refinements (CASS-inspired) 3. **Skill Artifact System** — Nanograph-native skill nodes with typed edges, SKILL.md-inspired schema, context-aware loading, effectiveness tracking 4. **CASS External Integration** — Optional cross-tool coding memory via external CASS query in agent context injection ## Dependencies - San-Ma (Milestone 1) must be complete — three-space separation is prerequisite for knowledge promotion and self-space skill storage - Additional dogfooding cycles needed before implementation begins ## Design Decisions (from research session 2026-03-06) - **CASS**: External tool, not a dependency. Kata queries it optionally; doesn't build session mining - **Meta Skill**: Adopt patterns (SKILL.md format, bandit suggestions, effectiveness tracking), build natively in nanograph - **TOON**: Build minimal TS encoder now. Port to Rust later. `--format toon` not `--robot` - **Cross-tool memory**: Not Kata's job to build, but agent context should include it when available - **Rust port triage**: Data model and workflows in TS. Performance-critical computation (graph algorithms, session mining, bandit engine) waits for Rust ## Research Sources - `tankyu/research/cass-memory-system-research-report.md` - `tankyu/research/meta-skill-research-report.md` - `tankyu/research/toon-rust-research-report.md` - `tankyu/research/ultimate-bug-scanner-research-report.md` - `tankyu/research/agentic-coding-flywheel-setup-research-report.md`
No due date•0/18 issues closedTransform .kata/ from a flat directory into a three-space architecture (self/, knowledge/, ops/) that structurally encodes data durability, enables single-root agent bootstrapping, unlocks cross-run knowledge queries, and prepares the foundation for nanograph integration. Vision doc: docs/vision/three-space-architecture.md
No due date•1/9 issues closed