kairos-chain 3.25.0 → 3.25.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 31f36e0972d3b7f0848a2a5334d18e5dce69ca4e9f507d0e8072ca9b263983b8
-  data.tar.gz: 3d4c0ff8590721b645088b386e2d3acfc7944fae62b774776a4fd2f4f65887ab
+  metadata.gz: 2ff49b5dba49e9c78161990be5bbf9bc94252542aeafe636b0c6cd424952ccff
+  data.tar.gz: 816f240e2cea9d045b4ac0c19c622792fe603dc49d2b454ec9409a8a6ab4f74a
 SHA512:
-  metadata.gz: 781cc48a6a9e55de327e2ca7cf4bee5d1e195dc9dc4c8f86e7998e36dc2d93ac301bd17c60efcacb0b63d4347d352c83c365d24cc7415e2f319f3a7276741c19
-  data.tar.gz: 7f91d5619741e422a6594bddb22a8bfba4bacf31de8681424e46a91c5a7ffa3b71d1d5ecaf21306a66680db0136efe252744e9dc9ae5be3446a8d50916b8e306
+  metadata.gz: aa98790fe6f4b71d1995b6d6a6822692a246c5aff24e3d99c9087ed44e78c9dc92ee19497628335862b1b6c93d156da514a7059bb275116d6e2ee62474e091a2
+  data.tar.gz: d50285be992138c811fb27bf9bf375648bbab801b71f37c9150d980996fc418da9e8bbd0bbaed6b506d19ecb187df932395f0a70f4501772b027c2bea8695581

data/CHANGELOG.md

@@ -4,6 +4,33 @@ All notable changes to the `kairos-chain` gem will be documented in this file.
 
 This project follows [Semantic Versioning](https://semver.org/).
 
+## [3.25.2] - 2026-05-07
+
+### Changed (L1 knowledge: reviewer evaluation feedback loop)
+
+- `multi_llm_review_workflow` § L2 Save Points: 各 review round 終了時に
+  per-reviewer observation (verdict, (a)/(b)/(c) breakdown, briefing-reaction
+  shift, anomalies) を `reviewer_evaluation_observation_<reviewer>_<date>`
+  prefix で L2 context に記録するよう明示。次回以降の
+  `multi_llm_reviewer_evaluation` refinement 用 sample 蓄積 channel として
+  workflow に組み込み。
+- `multi_llm_reviewer_evaluation` 末尾に "Refinement Source" section を追加。
+  上記 L2 context を refinement の source として明示することで L2 → L1
+  promotion loop を reviewer profile 自身に対しても閉じる (Prop 5
+  constitutive recording + Prop 6 incompleteness as driving force)。
+
+Surface 拡張なし: 既存セクションへの bullet 追加 + 新規 1 段落のみ。新 mechanism /
+新 field / 新 tool なし。
+
+## [3.25.1] - 2026-05-07
+
+### Changed (L1 knowledge: multi-LLM review)
+
+- `multi_llm_reviewer_evaluation` v1.2 → v1.3: harness memory に分散していた reviewer 性癖知識 (Codex 3 structural biases、Cursor vs Codex briefing-reaction data、Codex GPT-5.5 profile) を統合。新セクション "Reviewer Value-System Divergence" + (a)/(b)/(c) finding classification を追加。Convergence Rule を分類後ベース ((a)+(b) のみ blocking) に更新。Cost-Benefit を "Phase 1 baseline (5 reviewers)" にリネームし scope 明示。
+- `multi_llm_review_workflow`: Step 0 (mandatory `knowledge_get multi_llm_reviewer_evaluation`) と Step 0.5 (Design Direction Block for design / docs reviews) を追加。§ Convergence Rules と § Workflow Pattern step [4] を (a)/(b)/(c)-aware に整合。Step 0.5 block structure に invariant preface を追加 (anti-enumeration 整合)。
+
+設計の経緯と検証は self-review 2 round (Codex GPT-5.5 / Cursor Composer-2 / Claude CLI Opus 4.6 / Persona Team Opus 4.7) で実施。4/4 APPROVE / APPROVE WITH CHANGES、no REJECT。Phase 2 Case A (Context Graph review loop, 2026-05-04) で観察された value-system divergence を起点とし、KairosChain_2026 only の experimental briefing protocol (project CLAUDE.md) を operational extension として L1 化。
+
 ## [3.25.0] - 2026-05-07
 
 ### Added (Instruction mode projection)

data/bin/kairos-chain

@@ -348,6 +348,42 @@ when 'mode'
 
   mode_action = ARGV.shift || 'project'
 
+  if %w[-h --help help].include?(mode_action)
+    puts <<~HELP
+      Usage: kairos-chain mode <action> [--data-dir DIR]
+
+      Project the active instruction mode (Masa Mode, Tutorial Mode, ...)
+      to project-root CLAUDE.md via a managed @-import region. Required
+      for the mode body to reach Agent tool sub-agents (which do not
+      receive MCP `instructions`) and to bypass the harness truncation
+      cap on long mode bodies.
+
+      Actions:
+        project   Materialize the active mode body to .claude/kairos/
+                  instruction_mode.md and merge a marker region into
+                  project-root CLAUDE.md. Default action when no action
+                  is given. Idempotent — safe to re-run.
+        status    Print the current projection state (active mode name,
+                  version, artifact path/size, region presence, last
+                  projection time).
+        remove    Delete the projected artifact and remove the marker
+                  region from CLAUDE.md. Manifest is cleared.
+
+      Options:
+        --data-dir DIR   Override the .kairos/ data directory location.
+
+      Notes:
+        - The active mode is read from `instructions_mode` in
+          .kairos/skills/config.yml. Use `instructions_update` MCP tool
+          to change it; then re-run `mode project`.
+        - CLAUDE.md @-imports resolve at Claude Code session start;
+          you must restart Claude Code (`exit` then `claude`) for any
+          projection or removal to take effect.
+        - Body size policy: warn at >=150KB, refuse at >=256KB.
+    HELP
+    exit 0
+  end
+
   $LOAD_PATH.unshift File.expand_path('../lib', __dir__)
   require 'kairos_mcp'
 
@@ -484,6 +520,9 @@ OptionParser.new do |opts|
     puts "  init [DIR]          Initialize data directory with default templates"
     puts "  upgrade [--apply]   Check/apply template migrations after gem update"
     puts "  skillset <cmd>      Manage SkillSet plugins (list/install/enable/disable/remove/info)"
+    puts "  mode <action>       Project active instruction mode to CLAUDE.md (project/status/remove)"
+    puts ""
+    puts "Run a subcommand with -h for details, e.g. 'kairos-chain mode -h'."
     exit
   end
 end.parse!

data/lib/kairos_mcp/version.rb

@@ -1,4 +1,4 @@
 module KairosMcp
-  VERSION = "3.25.0"
+  VERSION = "3.25.2"
   CHANGELOG_URL = "https://github.com/masaomi/KairosChain_2026/blob/main/CHANGELOG.md"
 end

data/templates/knowledge/multi_llm_review_workflow/multi_llm_review_workflow.md

@@ -234,8 +234,10 @@ The user always has the final say.
          ├── outputs:  revised artifact + new review prompt
          └── L2 save:  consensus + revised artifact
          |
-[4] If 0 FAIL → proceed to next phase
-    If FAIL    → repeat from [2] with revised artifact
+[4] Classify findings as (a)/(b)/(c) per `multi_llm_reviewer_evaluation`
+    If no (a)/(b) blocking findings → proceed to next phase
+    If any (a)/(b) finding          → repeat from [2] with revised artifact
+    (c) findings are recorded as advisory; non-blocking
 ```
 
 ## Review Types
@@ -263,10 +265,21 @@ likely to be missed by a single LLM reviewing its own design. For per-model prof
 
 ## Convergence Rules
 
-- **3/4 APPROVE** (no REJECT) = proceed to next step
-- **Any REJECT or FAIL** = revise and re-review
-- **4/4 APPROVE** = highest confidence, proceed
-- Legacy 3-reviewer mode: 2/3 APPROVE = proceed
+The rule applies **after** orchestrator classifies each finding as (a)/(b)/(c) per
+`multi_llm_reviewer_evaluation` § Reviewer Value-System Divergence. Only (a)+(b)
+findings count toward the thresholds below; (c) findings are recorded as advisory
+and never block.
+
+- **3/4 APPROVE** (no (a)/(b) REJECT) = proceed to next step
+- **Any (a) or (b) REJECT or FAIL** = revise and re-review
+- **(c)-only REJECT** = record as advisory, non-blocking
+- **4/4 APPROVE** (no (a)/(b)) = highest confidence, proceed
+- Legacy 3-reviewer mode: 2/3 APPROVE (no (a)/(b)) = proceed
+- Codex REJECT with (a)/(b) findings + others APPROVE = likely real issue, investigate before overriding
+- Codex REJECT with only (c) findings = expected per Codex value-system divergence; non-blocking
+
+For normative detail and the underlying classification, see
+`multi_llm_reviewer_evaluation` § Convergence Rule (Updated).
 
 ### Consensus Patterns
 
@@ -331,6 +344,14 @@ Save to L2 context at these moments:
 - After design/implementation complete (before review)
 - After synthesis of reviews (revised version)
 - After final convergence (implementation-ready / merge-ready)
+- **After each review round**: capture per-reviewer observations — verdict,
+  (a)/(b)/(c) classification breakdown, briefing-reaction shift (did the
+  reviewer change verdict after Step 0.5 design direction?), anomalies
+  (off-pattern findings, format failures, refusal). Tag context name with
+  prefix `reviewer_evaluation_observation_<reviewer>_<date>` so future
+  refinement of `multi_llm_reviewer_evaluation` can sample these records
+  systematically. This closes the L2→L1 promotion loop for reviewer
+  profiles themselves.
 
 ---
 

data/templates/knowledge/multi_llm_reviewer_evaluation/multi_llm_reviewer_evaluation.md

@@ -271,7 +271,8 @@ Deployment:         Composer-2 or Cursor GPT-5.4
 | Reviewer | Summary |
 |----------|---------|
 | Claude Opus 4.6 | Guardian of design. Finds security threats and novel architectural alternatives |
-| Codex GPT-5.4 | Strictest judge. Last to approve, but APPROVE = highest confidence signal |
+| Codex GPT-5.4 | Strictest judge. Classify findings (a)/(b)/(c) before treating REJECT as blocking; APPROVE is a strong signal **when reachable**, not a mandatory gate (see Phase 2 Case A caveat) |
+| Codex GPT-5.5 | Stricter sibling of 5.4. Same value-system divergence (3 biases); apply the same classification discipline |
 | Cursor Premium | Implementation craftsman. Bug hunter for concurrency and resource management |
 | Composer-2 | Fastest pragmatist. First to determine if something is deployable |
 | Cursor GPT-5.4 | Binary sword. Clear approve-or-reject, strictest on test coverage |
@@ -292,3 +293,12 @@ Deployment:         Composer-2 or Cursor GPT-5.4
 5. Some REJECTs reflect the reviewer's value system, not the artifact. The (a)/(b)/(c)
    classification (see § Reviewer Value-System Divergence) is required to separate
    blocking signal from advisory noise. Codex models in particular require this lens.
+
+## Refinement Source
+
+Profiles in this knowledge are refined from accumulated L2 contexts named with prefix
+`reviewer_evaluation_observation_<reviewer>_<date>`, recorded after each multi-LLM
+review round per `multi_llm_review_workflow` § L2 Save Points. When updating this
+file, sample those records to revise per-reviewer profiles, Strength Matrix entries,
+Cost-Benefit ratings, and the value-system divergence section. This closes the
+L2 → L1 promotion loop for reviewer profiles themselves.

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: kairos-chain
 version: !ruby/object:Gem::Version
-  version: 3.25.0
+  version: 3.25.2
 platform: ruby
 authors:
 - Masaomi Hatakeyama