v1.1.0 — The Reasoning Depth Release
Focused upgrade: deeper review reasoning, resumable builds, and systematic coverage mapping.
Full v1.1.0 Workflow
New Skills
meow:elicit — Structured second-pass reasoning
Post-verdict deeper analysis using 8 elicitation methods. Invoked after a reviewer verdict to push harder on borderline dimensions, surface hidden assumptions, or stress-test a PASS decision before Gate 2.
The 8 methods probe from different angles: devil's advocate, steelman, inversion, pre-mortem, boundary probing, assumption surfacing, counterfactual, and first-principles. Each method is independent — elicitation selects the subset most relevant to the verdict's weak points.
meow:validate-plan — 8-dimension plan quality check
Runs automatically after the planner produces a plan file, before Gate 1 is presented. Checks 8 dimensions: problem clarity, binary success criteria, explicit out-of-scope, pattern fit, actionable risks, clean bead boundaries, realistic effort, and testable acceptance criteria.
A plan that fails any dimension is revised before the human sees it. Gate 1 is only presented after validate-plan passes.
meow:validate-plan reference →
meow:nyquist — Test-to-requirement coverage mapping
Maps written tests to acceptance criteria from the plan file. Ensures that every binary criterion has at least one test targeting it — before Phase 3 begins. Named after Nyquist sampling theorem: you need sufficient test density to faithfully represent the requirement space.
Run at the end of Phase 2 (Test RED) to confirm coverage before handing off to the developer.
Enhanced Review Pipeline
Scout integration
meow:scout runs optionally before the main review to detect edge cases, unusual patterns, and context the parallel reviewers might miss. Scout findings are injected into the review context as a pre-loaded brief.
Elicitation hook
After the reviewer issues a verdict, meow:elicit can be invoked for a structured second pass. This is particularly useful when a dimension produced WARN — elicitation either upgrades WARN to PASS (with evidence) or downgrades it to FAIL (with specific findings).
Execution Resilience
Beads pattern for COMPLEX builds
COMPLEX tasks (5+ files) are decomposed into beads — atomic, independently committable work units. Each bead covers a bounded set of files, has a clear done state, and produces one conventional commit.
If the session is interrupted, the developer resumes from the last uncommitted bead. No work is duplicated or lost.
bead-01 → commit → bead-02 → commit → bead-03 → ...Bead template for plan decomposition
The planner includes a bead section in COMPLEX plan files. Each bead entry specifies: files owned, acceptance check, and commit message template. Developers follow the bead list sequentially without re-interpreting scope at each step.
Agent System
Subagent Status Protocol
Every subagent MUST end responses with a structured status block. Four statuses, deterministic routing:
| Status | Meaning |
|---|---|
| DONE | Task completed successfully |
| DONE_WITH_CONCERNS | Completed, flagged doubts |
| BLOCKED | Cannot complete — something must change before retry |
| NEEDS_CONTEXT | Missing information — provide it, then re-dispatch |
Sub-agent type classification
Agents are now classified as Core (pipeline, sequential) or Support (on-demand, spawnable). The agents index includes a Type column. See Agents Overview.
AGENTS_INDEX.md
Centralized registry at .claude/agents/AGENTS_INDEX.md with columns: Agent file, Type, Role, Source, Workflow phases, Auto-activate, CE version, Last improved.
SKILLS_INDEX.md
Centralized skill registry at .claude/agents/SKILLS_INDEX.md — single source of truth for skill discovery, dependencies, and activation conditions.
Documentation
Enforcement mechanism matrix in RULES_INDEX
RULES_INDEX.md now includes a full enforcement mechanism matrix: each rule mapped to its mechanism (Behavioral / Hook / Data), whether it can be overridden, and the exception condition.
Quick-start skill selection guide
New section in the skills docs: a decision table for selecting the right skill given task type, complexity, and phase. Replaces the need to read every SKILL.md before knowing what to use.
Delegation checklist in orchestration rules
orchestration-rules.md now includes a pre-delegation checklist (5 items) and a delegation prompt template. Prevents the most common subagent failure: missing file ownership or acceptance criteria in the prompt.