v2.0.0 — The Leverage Release
The biggest MeowKit update yet. Extracted high-leverage patterns from ECC's 38-agent, 156-skill ecosystem and adapted them for MeowKit's disciplined workflow. Result: 5 new skills, 17 reference merges across 10 skills, hook profiling, naming cleanup, and rule relaxations — without adding agents or compromising hard gates.
Thesis: Keep the discipline, add leverage.
Highlights
5 New Skills
| Skill | Purpose | Phase |
|---|---|---|
meow:decision-framework | Operational decision architecture: triage, escalation, case management | 1 (Plan) |
meow:verify | Unified verification: build→lint→test→type-check→coverage | 3→4 transition |
meow:api-design | REST/GraphQL API design patterns | 1 (Plan) |
meow:build-fix | Build error triage with language-specific fix references | 3 (Build) |
meow:database | Schema design, migrations, query optimization (PostgreSQL primary) | 1/3 |
17 Reference Merges Across 10 Skills
Existing skills got deeper with new JIT-loaded references:
| Skill | New References |
|---|---|
| meow:investigate | RCA method selection (5 Whys / Ishikawa / 8D / Fault Tree), anti-patterns |
| meow:plan-creator | Ops metrics design, cold-start context briefs, plan mutation protocol, worked example |
| meow:qa | 4-phase browser QA checklist (smoke → interaction → visual → a11y) |
| meow:agent-detector | Token budget depth levels (25/50/75/100%) |
| meow:office-hours | Founder Review + User Journey Audit modes |
| meow:typescript | Prioritized TS/JS review checklist (CRITICAL → MEDIUM) |
| meow:cook | Loop safety protocol (stall detection, cost drift, escalation) |
| meow:review | Iterative evaluation protocol (max 3 passes for high-stakes code) |
| meow:frontend-design | Anti-slop directives (avoid AI-generated UI cliches) |
| meow:testing | E2E best practices (Agent Browser, POM, flaky quarantine) |
Hook Runtime Profiling
New MEOW_HOOK_PROFILE environment variable controls hook intensity:
| Profile | Hooks Active | Use When |
|---|---|---|
strict | All 10 hooks | COMPLEX tasks, security-critical |
standard | 8 hooks (skip cost-meter, post-session) | Default — everyday development |
fast | 3 hooks (gate-enforcement, privacy-block, context-loader only) | Rapid iteration, prototyping |
# Set in .env or shell
MEOW_HOOK_PROFILE=fastSafety-critical hooks (gate-enforcement, privacy-block) never skip, regardless of profile.
Naming Cleanup
| Old | New | Status |
|---|---|---|
| meow:shipping | meow:ship | Merged (redirect in place) |
| meow:documentation | meow:document-release | Merged (redirect in place) |
| meow:debug | meow:investigate | Merged (redirect in place) |
Redirects stay for 2 releases. Old invocations show deprecation notice.
Rule Relaxations
MICRO-TASK TDD Exemption — Non-production code <30 lines classified as MICRO-TASK by orchestrator is exempt from TDD. Production code always requires TDD regardless of size. (Distinct from TRIVIAL which is cosmetic-only.)
Staged Parallel Mode — When strict zero-file-overlap is impractical, split overlapping files sequentially while non-overlapping work runs in parallel. Same max-3-agent limit.
Memory Capture Enhancement — Budget extended from 2 min / 3 markers to 5 min / 5 markers. CRITICAL/SECURITY markers processed regardless of age. New --capture-all override.
Mandatory Simplification Step
meow:cook workflow now includes a mandatory meow:simplify pass between Phase 3 (Build) and Phase 4 (Review). Catches over-engineering before review.
Proactive Learning Observer
New learning-observer.sh PostToolUse hook detects session patterns:
- File edited 3+ times → churn pattern logged
- Feeds into
post-session.shretroactive capture
Active in strict and standard profiles.
External Integrations
Three new skills for connecting MeowKit to external design and project management tools:
| Skill | Purpose |
|---|---|
meow:intake | Tool-agnostic ticket/PRD analysis: completeness scoring, product area classification, RCA, media handling |
meow:jira | Jira execution via Atlassian MCP: create, search, update, transition, link, sprint management |
meow:figma | Figma design analysis via Figma MCP: extract design context, translate to code, extract tokens |
Workflow: meow:intake analyzes → meow:jira executes Jira actions → meow:figma extracts design context → meow:cook implements.
See PRD Intake Automation and Tickets with Media for full workflows.
Migration
All changes are non-breaking. Deprecated skill names have redirect SKILL.md files. New MEOW_HOOK_PROFILE defaults to standard (matches pre-v2.0 behavior). Upgrade with:
npx mewkit init # re-run to get new skills + references