Henkaten Council plugin for Claude Code — governance layer that orchestrates change-point detection, autonomous decision audit, and continuous improvement across trine-eval sprints.
Henkaten Council Orchestrator. Level 4 agent responsible for coordinating the full council review cycle: dispatching worker agents, honoring andon signals, managing the dynamic autonomy floor, spot-checking evidence, writing decision-log entries, and presenting major corrections via the nemawashi walkthrough. The orchestrator is the conductor — it sees the user's full input and sequences the council's work within bounded, supervised workflows.
Coherence and drift reviewer. Proposal-only agent. Reads sprint results, eval reports, spec, features, and sprint plan to assess whether the implementation is coherent with the plan and flag architectural drift, dependency health, and risk. Produces evidence-cited proposals; does not modify any files.
Use this skill to run the council fan-out sequence (henkaten check, agent fan-out, decision logging, retrospective) for the next harness sprint, after the user runs `/trine-eval:harness-sprint`. Invoked as the `council-autorun` slash command under the `henkaten-council` plugin. Runs the full ten-step governance loop defined in §8.2 of the Phase-0 v2 proposal: pre-sprint henkaten check, yokoten review, trine-eval delegation, council fan-out, course correction, decision logging, halt condition evaluation, context compaction, mini retrospective, and next-sprint trigger logic.
Use this skill to run on-demand henkaten (change-point) detection against the current project state. Classifies detected changes by 4M axis and change_origin (active vs passive), applies configurable sensitivity thresholds (conservative / balanced / sensitive), and writes detected records to .council/henka-register.jsonl via scripts/append-henka.py.
Use this skill to run an on-demand, user-invoked reflection workshop on a chosen topic. Operates in jishuken mode: produces Reflection Notes, Open Questions, and Hypotheses. This is a reflection-only skill — it does NOT produce standard-work proposals and does NOT propose changes to standard-work.json. Floor restoration is exclusively handled by council-review --restore-autonomy, not by this skill.
Use this skill to bootstrap a new henkaten-council governance baseline in a trine-eval project. Invoke it once per project before running any council review cycle. It creates the complete .council/ directory structure, seeds all required state files, writes the governance signal to .harness/config.json, and delegates initial sprint planning to /trine-eval:harness-kickoff (or /trine-eval:harness-sprint for subsequent sprints).
Use this skill to run the per-sprint mini retrospective automatically after every sprint completes. Invoked by council-autorun at Step 1H; runs in ≤30s wall-clock time with no user input required. Capture-only: produces Learning Points and Pattern Observations. No standard-work proposals are permitted in mini mode.
Matches all tools
Hooks run on every tool call, not just specific ones
Executes bash commands
Hook triggers when Bash tool is used
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Modifies files
Hook triggers on file write and edit operations
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Uses power tools
Uses Bash, Write, or Edit tools
A Claude Code plugin that adds a continuous-improvement governance layer on top of any project managed by the trine-eval harness. The council detects change points (henkaten), audits autonomous decisions, manages a dynamic autonomy floor, and enforces the andon protocol across sprints.
A note on the name. henka (変化) means "change"; henkaten (変化点) is the specific Toyota Production System term for a "change point" — the concept this plugin is built around. The repo got the shorter
henka-councilfor URL brevity, but the plugin identifier, the skill namespace (/henkaten-council:...), and every in-code reference use the fullhenkaten-council. The contraction is deliberate and documented indocs/phase-0-proposal.md. Internally the plugin also distinguishes henkaten (変化点 — passive, emerged) from henkoten (変更点 — active, deliberately initiated) on every change-point record, but uses "henkaten" as the umbrella name.
The henkaten-council plugin wires a council of specialized agents into the
Claude Code skill system. Each sprint, the council:
trine-eval harness via
/trine-eval:harness-sprint.council/trine-eval ≥ 0.3.0 installed and initialized in your target project
(run /trine-eval:harness-kickoff first if the .harness/ directory does
not yet exist)This repository ships its own single-plugin marketplace manifest
(.claude-plugin/marketplace.json). In Claude Code, add the repo as a
marketplace, then install the plugin from it:
/plugin marketplace add ats-kinoshita-iso/henka-council
/plugin install henkaten-council@henkaten-council
Claude Code will fetch the plugin, auto-discover the skills and agents
under skills/ and agents/, auto-register the hooks declared in
hooks/hooks.json, and load CLAUDE.md automatically for all future
sessions in the project.
Clone this repository and add it as a local marketplace:
git clone https://github.com/ats-kinoshita-iso/henka-council.git
/plugin marketplace add ./henka-council
/plugin install henkaten-council@henkaten-council
After installation, verify the plugin is active:
/plugin list
You should see henkaten-council in the output with status active.
The council's enforcement hooks must be registered with Claude Code's hook
system before the council can prevent append-only log overwrites or enforce
the reversibility policy. The hooks live in hooks/ (Bash, for Linux/macOS)
and hooks/win/ (PowerShell, for Windows).
As of v0.1.2: hooks auto-register from
hooks/hooks.jsonwhen the plugin is installed via the marketplace flow (Option A above). The kickoff skill's Step 1d.0 detects this case and the manual snippet below is then only a fallback — needed for direct-path installs on Claude Code builds that don't read plugin-levelhooks.json, or for pure-PowerShell Windows hosts without bash.
When you run /henkaten-council:council-kickoff for the first time, the skill
performs a hook installation self-check and refuses to proceed past Step 1d
if neither auto-registration nor manual registration covers all four required
hooks.
The full registration snippet is in
skills/council-kickoff/SKILL.md under
Step 1d.1 (Linux/macOS) and Step 1d.2 (Windows). The short version:
Add this hooks block to your target project's .claude/settings.local.json:
"hooks": {
"PreToolUse": [
{ "matcher": "Write|Edit",
"hooks": [{ "type": "command",
"command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/enforce-append-only.sh" }] },
{ "matcher": "Bash",
"hooks": [{ "type": "command",
"command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/enforce-reversibility.sh" }] }
],
"PostToolUse": [
{ "matcher": "*",
"hooks": [{ "type": "command",
"command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/log-tool-call.sh" }] }
],
"Stop": [
{ "hooks": [{ "type": "command",
"command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/session-stopped-marker.sh" }] }
]
}
npx claudepluginhub ats-kinoshita-iso/henka-council --plugin henkaten-councilEvaluation framework skills for designing scoring rubrics, running structured evaluations on LLM outputs, and comparing candidate outputs to recommend a winner.
trine-eval: Planner-Generator-Evaluator harness for eval-driven development across web/RAG/CLI/API projects, eval-harness methodology audits (meta layer), and harness-build agent-runtime conformance (runtime layer)
Analyzes and rewrites prompts for Claude Code, applying structured prompt engineering patterns to produce clearer, more effective instructions.
Anthropic's official development skills for Claude API integration, MCP server building, skill creation, web artifact building, and browser-based testing.
Skills for designing and evaluating multi-agent systems: orchestrator/worker decomposition, output quality review, and self-improving evaluator/optimizer loops.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Comprehensive startup business analysis with market sizing (TAM/SAM/SOM), financial modeling, team planning, and strategic research
v9.44.1 — Patch release for Gemini environment/version detection and qwen auth gating. Run /octo:setup.
Permanent coding companion for Claude Code — survives any update. MCP-based terminal pet with ASCII art, stats, reactions, and personality.
Complete creative writing suite with 10 specialized agents covering the full writing process: research gathering, character development, story architecture, world-building, dialogue coaching, editing/review, outlining, content strategy, believability auditing, and prose style/voice analysis. Includes genre-specific guides, templates, and quality checklists.
Comprehensive .NET development skills for modern C#, ASP.NET, MAUI, Blazor, Aspire, EF Core, Native AOT, testing, security, performance optimization, CI/CD, and cloud-native applications