@hongmaple0820/scale-engine 0.29.0 → 0.33.0

package/README.en.md

@@ -35,6 +35,8 @@ This is not a claim that SCALE replaces human judgment. It is the first testable
 
 The near-term target is `0.28.0` as a usable closed-loop enhancement: connect `ai-os plan`, `ai-os run`, verification recommendations, failure learning, dashboard, benchmark, migration, and adoption into one verifiable loop. The long-range target is an AI Engineering OS beta in 8-12 weeks, a stable governance runtime in 3-6 months, and a cross-agent engineering operating layer in 6-12 months. See the full roadmap in [AI Engineering OS Strategic Positioning](docs/AI_ENGINEERING_OS_POSITIONING.md).
 
+The first 0.30.0 governance-maturity slice adds Evaluator Intelligence and Tool Strategy Planner to the AI OS runtime. `scale ai-os plan` now detects reasoning-heavy architecture, root-cause, security, and release work, then adds critique, threat-model, release-readiness, and uncertainty decision-log gates to the adaptive workflow. It also turns skill/artifact/verification steps into a cost, retry, fallback, side-effect, and evidence graph. `scale ai-os status` surfaces evaluator gate count, uncertainty, tool-strategy cost, and fallback coverage so reviewers can see whether reasoning and tool risks were governed instead of hidden in prose.
+
 The current 0.27.0 beta runtime now includes the controlled run entry point: `scale ai-os run --dry-run` reuses the unified plan, produces execution steps, evidence requirements, next actions, and writes the run report to `.scale/ai-os/runs/`. When real verification is required, use guarded mode with explicit `--verify` commands. Commands run through the safe runner by default and are recorded as runtime evidence; failed verification returns a `blocked` JSON report and a non-zero CLI exit code.
 
 ```bash

package/README.md

@@ -19,6 +19,8 @@ npm：https://www.npmjs.com/package/@hongmaple0820/scale-engine
 
 ## 0.27.0 AI OS Runtime
 
+> 0.30.0 治理成熟度预览：AI OS Runtime 已加入 Evaluator Intelligence 和 Tool Strategy Planner。`scale ai-os plan` 会识别架构、根因、安全、发版等推理风险任务，并把 architecture critique、root-cause review、security threat model、release readiness 和 uncertainty decision log 加入 adaptive workflow；同时把 skill、artifact、verification 步骤编译成 cost、retry、fallback、side-effect 和 evidence graph。`scale ai-os status` 会展示 evaluator gate、uncertainty、tool strategy cost 和 fallback coverage，让评审者看到推理风险和工具风险是否被门禁治理，而不是只藏在文字说明里。
+
 0.27.0 把战略方向落成了一个可执行入口：`scale ai-os plan`。它会在一次命令里同时生成风险治理模式、Context Compiler 预算结果、Memory Provider 召回结果、Skill Routing 执行计划和 Governance ROI，让 Agent 在开始任务前就知道应该加载什么上下文、调用什么能力、补什么证据、哪些风险会升级门禁。
 
 ```bash

package/dist/api/cli.js

@@ -3240,6 +3240,8 @@ const aiOsStatusCommand = defineCommand({
             console.log(`  Doctor: ${report.doctor.status}`);
             console.log(`  Intelligence: ${report.intelligence.status} (${report.intelligence.summary.ready} ready, ${report.intelligence.summary.warning} warning, ${report.intelligence.summary.blocked} blocked)`);
             console.log(`  Context risk: ${report.intelligence.summary.contextQuality.compressionRisk}; omitted ${report.intelligence.summary.contextQuality.omittedSections} section(s), evidence warnings ${report.intelligence.summary.contextQuality.evidenceLossWarnings.length}`);
+            console.log(`  Evaluator gates: ${report.intelligence.summary.evaluatorQuality.requiredGates}; uncertainty ${report.intelligence.summary.evaluatorQuality.averageUncertainty}`);
+            console.log(`  Tool strategy: ${report.intelligence.summary.toolStrategyQuality.totalSteps} step(s), cost ${report.intelligence.summary.toolStrategyQuality.estimatedCostUnits}, fallback ${report.intelligence.summary.toolStrategyQuality.fallbackCoverage}`);
             for (const signal of report.intelligence.signals)
                 console.log(`  [${signal.status}] ${signal.id}: ${signal.summary}`);
             for (const check of report.checks)
@@ -3263,6 +3265,8 @@ const aiOsStatusCommand = defineCommand({
             console.log(`  Doctor: ${report.doctor.status}`);
             console.log(`  Intelligence: ${report.intelligence.status} (${report.intelligence.summary.ready} ready, ${report.intelligence.summary.warning} warning, ${report.intelligence.summary.blocked} blocked)`);
             console.log(`  Context risk: ${report.intelligence.summary.contextQuality.compressionRisk}; omitted ${report.intelligence.summary.contextQuality.omittedSections} section(s), evidence warnings ${report.intelligence.summary.contextQuality.evidenceLossWarnings.length}`);
+            console.log(`  Evaluator gates: ${report.intelligence.summary.evaluatorQuality.requiredGates}; uncertainty ${report.intelligence.summary.evaluatorQuality.averageUncertainty}`);
+            console.log(`  Tool strategy: ${report.intelligence.summary.toolStrategyQuality.totalSteps} step(s), cost ${report.intelligence.summary.toolStrategyQuality.estimatedCostUnits}, fallback ${report.intelligence.summary.toolStrategyQuality.fallbackCoverage}`);
             for (const signal of report.intelligence.signals)
                 console.log(`  [${signal.status}] ${signal.id}: ${signal.summary}`);
             for (const check of report.checks)