fix(plugin): add explicit Agent tool warning against Bash CLI dispatch (#189)

* fix(plugin): add explicit Agent tool warning against Bash CLI dispatch

Skills now clearly instruct the host to use the built-in Agent tool
instead of shelling out to `claude` CLI for agent dispatch. Adds test
assertion for the new contract language.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore(plugin): rebuild dist artifacts

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: onlycastle

packages/plugin/dist/chunk-NH4BKYT6.js packages/plugin/dist/chunk-QFCYET5Y.jspackages/plugin/dist/chunk-NH4BKYT6.js renamed to packages/plugin/dist/chunk-QFCYET5Y.js

@@ -27560,6 +27560,9 @@ var ParsedMessageSchema = external_exports2.object({
   role: external_exports2.enum(["user", "assistant"]),
   timestamp: external_exports2.string(),
   content: external_exports2.string(),
+  isMeta: external_exports2.boolean().optional(),
+  sourceToolUseID: external_exports2.string().optional(),
+  toolUseResult: external_exports2.unknown().optional(),
   toolCalls: external_exports2.array(ToolCallSchema).optional(),
   tokenUsage: external_exports2.object({
     input: external_exports2.number().int().min(0),
@@ -27686,12 +27689,18 @@ var Phase1OutputSchema = external_exports2.object({
 
 // ../shared/dist/schemas/deterministic-scores.js
 var DeterministicScoresSchema = external_exports2.object({
-  contextEfficiency: external_exports2.number().min(0).max(100),
-  sessionOutcome: external_exports2.number().min(0).max(100),
-  thinkingQuality: external_exports2.number().min(0).max(100),
-  learningBehavior: external_exports2.number().min(0).max(100),
-  communicationPatterns: external_exports2.number().min(0).max(100),
-  controlScore: external_exports2.number().min(0).max(100)
+  aiPartnership: external_exports2.number().min(0).max(100),
+  sessionCraft: external_exports2.number().min(0).max(100),
+  toolMastery: external_exports2.number().min(0).max(100),
+  skillResilience: external_exports2.number().min(0).max(100),
+  sessionMastery: external_exports2.number().min(0).max(100),
+  controlScore: external_exports2.number().min(0).max(100),
+  // Legacy fields — present on old runs, not computed for new
+  contextEfficiency: external_exports2.number().min(0).max(100).optional(),
+  sessionOutcome: external_exports2.number().min(0).max(100).optional(),
+  thinkingQuality: external_exports2.number().min(0).max(100).optional(),
+  learningBehavior: external_exports2.number().min(0).max(100).optional(),
+  communicationPatterns: external_exports2.number().min(0).max(100).optional()
 });
 var CodingStyleTypeSchema = external_exports2.enum([
   "architect",
@@ -27740,6 +27749,12 @@ var DomainGrowthAreaSchema = external_exports2.object({
 });
 var DomainResultSchema = external_exports2.object({
   domain: external_exports2.enum([
+    "aiPartnership",
+    "sessionCraft",
+    "toolMastery",
+    "skillResilience",
+    "sessionMastery",
+    // Legacy domains accepted for backward compat with old runs
     "thinkingQuality",
     "communicationPatterns",
     "learningBehavior",
@@ -27885,12 +27900,17 @@ var STAGE_NAMES = [
   "evidenceVerification",
   "contentWriter",
   "translator",
+  // 5-dimension extractors (v2)
+  "extractAiPartnership",
+  "extractSessionCraft",
+  "extractToolMastery",
+  "extractSkillResilience",
+  "extractSessionMastery",
+  // Legacy extractors (old runs only)
   "extractAiCollaboration",
   "extractContextEngineering",
-  "extractToolMastery",
   "extractBurnoutRisk",
-  "extractAiControl",
-  "extractSkillResilience"
+  "extractAiControl"
 ];
 var STAGE_SCHEMAS = {
   sessionSummaries: SessionSummaryBatchSchema,
@@ -27900,12 +27920,17 @@ var STAGE_SCHEMAS = {
   evidenceVerification: EvidenceVerificationOutputSchema,
   contentWriter: ContentWriterOutputSchema,
   translator: TranslatorOutputSchema,
+  // 5-dimension extractors (v2)
+  extractAiPartnership: DimensionExtractionSchema,
+  extractSessionCraft: DimensionExtractionSchema,
+  extractToolMastery: DimensionExtractionSchema,
+  extractSkillResilience: DimensionExtractionSchema,
+  extractSessionMastery: DimensionExtractionSchema,
+  // Legacy extractors (old runs)
   extractAiCollaboration: DimensionExtractionSchema,
   extractContextEngineering: DimensionExtractionSchema,
-  extractToolMastery: DimensionExtractionSchema,
   extractBurnoutRisk: DimensionExtractionSchema,
-  extractAiControl: DimensionExtractionSchema,
-  extractSkillResilience: DimensionExtractionSchema
+  extractAiControl: DimensionExtractionSchema
 };
 
 // ../shared/dist/schemas/analysis-run.js
@@ -28336,6 +28361,96 @@ var SessionOutcomeOutputSchema = external_exports2.object({
   growthAreas: external_exports2.array(WorkerGrowthSchema).optional(),
   referencedInsights: external_exports2.array(ReferencedInsightSchema).optional()
 });
+var AiPartnershipOutputSchema = external_exports2.object({
+  // Planning Dimension (from ThinkingQuality)
+  planningHabits: external_exports2.array(PlanningHabitSchema),
+  planQualityScore: external_exports2.number().min(0).max(100),
+  multitaskingPattern: MultitaskingPatternSchema.optional(),
+  // Critical Thinking Dimension (from ThinkingQuality)
+  verificationBehavior: VerificationBehaviorSchema,
+  criticalThinkingMoments: external_exports2.array(CriticalThinkingMomentSchema),
+  verificationAntiPatterns: external_exports2.array(DetectedAntiPatternSchema),
+  // AI Control Dimension (from SessionOutcome)
+  sessionAnalyses: external_exports2.array(SessionAnalysisSchema),
+  overallSuccessRate: external_exports2.number().min(0).max(100),
+  goalDistribution: external_exports2.array(GoalDistributionItemSchema),
+  frictionSummary: external_exports2.array(FrictionSummaryItemSchema),
+  // Overall Scores
+  overallAiPartnershipScore: external_exports2.number().min(0).max(100),
+  confidenceScore: external_exports2.number().min(0).max(1),
+  summary: external_exports2.string().optional(),
+  // Domain-specific Strengths & Growth Areas
+  strengths: external_exports2.array(WorkerStrengthSchema).optional(),
+  growthAreas: external_exports2.array(WorkerGrowthSchema).optional(),
+  referencedInsights: external_exports2.array(ReferencedInsightSchema).optional()
+});
+var SessionCraftOutputSchema = external_exports2.object({
+  // Context Efficiency Dimension
+  contextUsagePatterns: external_exports2.array(ContextUsagePatternSchema),
+  inefficiencyPatterns: external_exports2.array(InefficiencySchema),
+  promptLengthTrends: external_exports2.array(PromptLengthTrendSchema),
+  avgContextFillPercent: external_exports2.number().min(0).max(100),
+  // Burnout / Learning Dimension
+  repeatedMistakePatterns: external_exports2.array(RepeatedMistakePatternSchema),
+  knowledgeGaps: external_exports2.array(KnowledgeGapItemSchema),
+  // Insights
+  topInsights: external_exports2.array(external_exports2.string()).max(3),
+  // Overall Scores
+  overallSessionCraftScore: external_exports2.number().min(0).max(100),
+  confidenceScore: external_exports2.number().min(0).max(1),
+  summary: external_exports2.string().optional(),
+  // Domain-specific Strengths & Growth Areas
+  strengths: external_exports2.array(WorkerStrengthSchema).optional(),
+  growthAreas: external_exports2.array(WorkerGrowthSchema).optional(),
+  referencedInsights: external_exports2.array(ReferencedInsightSchema).optional()
+});
+var AbsenceIndicatorSchema = external_exports2.object({
+  /** Name of the anti-pattern being checked */
+  pattern: external_exports2.string(),
+  /** Whether the anti-pattern was observed */
+  present: external_exports2.boolean(),
+  /** Number of occurrences (0 = absent) */
+  occurrenceCount: external_exports2.number().int().min(0),
+  /** Total sessions checked */
+  sessionsChecked: external_exports2.number().int().min(1),
+  /** Score contribution: 0 (always present) to 100 (always absent) */
+  absenceScore: external_exports2.number().min(0).max(100),
+  /** Evidence of presence (empty if absent — which is good) */
+  evidence: external_exports2.array(EvidenceItemSchema),
+  /**
+   * Expert interpretation:
+   * - 'internalized': absence indicates mastery (skill no longer needs scaffolding)
+   * - 'not_applicable': pattern not relevant for this developer's workflow
+   * - 'concerning': absence may indicate gap (e.g., never verifies output)
+   */
+  interpretation: external_exports2.enum(["internalized", "not_applicable", "concerning"])
+});
+var SessionCleanlinessSchema = external_exports2.object({
+  sessionId: external_exports2.string(),
+  cleanlinessScore: external_exports2.number().min(0).max(100),
+  antiPatternCount: external_exports2.number().int().min(0),
+  /** Indicates expert-level session control */
+  isCleanSession: external_exports2.boolean()
+});
+var SessionMasteryOutputSchema = external_exports2.object({
+  // Absence Indicators
+  absenceIndicators: external_exports2.array(AbsenceIndicatorSchema),
+  // Session Cleanliness
+  sessionCleanliness: external_exports2.array(SessionCleanlinessSchema),
+  cleanSessionPercentage: external_exports2.number().min(0).max(100),
+  // Expert Differentiation
+  scaffoldingDependencyScore: external_exports2.number().min(0).max(100),
+  internalizedSkillSignals: external_exports2.array(external_exports2.string()),
+  expertBehaviorIndicators: external_exports2.array(external_exports2.string()),
+  // Overall Scores
+  overallSessionMasteryScore: external_exports2.number().min(0).max(100),
+  confidenceScore: external_exports2.number().min(0).max(1),
+  summary: external_exports2.string().optional(),
+  // Domain-specific Strengths & Growth Areas
+  strengths: external_exports2.array(WorkerStrengthSchema).optional(),
+  growthAreas: external_exports2.array(WorkerGrowthSchema).optional(),
+  referencedInsights: external_exports2.array(ReferencedInsightSchema).optional()
+});
 
 // ../shared/dist/evaluation/canonical-analysis.js
 var DOMAIN_TO_EVALUATION_DIMENSION = {
@@ -29026,14 +29141,56 @@ function scoreControl(metrics) {
   const commandSignal = Math.min(uniqueCommands * 10 + totalCommands * 2, 100);
   return clampScore(rejectionSignal * 0.25 + questionSignal * 0.25 + lengthSignal * 0.25 + commandSignal * 0.25);
 }
+function scoreAiPartnership(metrics) {
+  const thinking = scoreThinkingQuality(metrics);
+  const outcome = scoreSessionOutcome(metrics);
+  const control = scoreControl(metrics);
+  return clampScore(thinking * 0.4 + outcome * 0.35 + control * 0.25);
+}
+function scoreSessionCraft(metrics) {
+  const efficiency = scoreContextEfficiency(metrics);
+  const learning = scoreLearningBehavior(metrics);
+  return clampScore(efficiency * 0.55 + learning * 0.45);
+}
+function scoreSkillResilience(metrics) {
+  const totalSessions = Math.max(metrics.totalSessions, 1);
+  const totalUtterances = Math.max(metrics.totalDeveloperUtterances, 1);
+  const shortSessions = metrics.sessionHints?.shortSessions ?? 0;
+  const shortRatio = shortSessions / totalSessions;
+  const coldStartScore = bellCurveScore(shortRatio * 100, 20, 50, 3e-3);
+  const bareRetryRate = (metrics.frictionSignals?.bareRetryAfterErrorCount ?? 0) / totalUtterances;
+  const recoveryScore = invertedScale(bareRetryRate * 200);
+  const slashCmds = metrics.slashCommandCounts ?? {};
+  const uniqueCommands = Object.keys(slashCmds).length;
+  const diversityScore = Math.min(uniqueCommands * 12 + 20, 100);
+  return clampScore(coldStartScore * 0.3 + recoveryScore * 0.4 + diversityScore * 0.3);
+}
+function scoreSessionMastery(metrics) {
+  const totalSessions = Math.max(metrics.totalSessions, 1);
+  const totalUtterances = Math.max(metrics.totalDeveloperUtterances, 1);
+  const friction = metrics.frictionSignals;
+  const excessiveIterationRate = (friction?.excessiveIterationSessions ?? 0) / totalSessions;
+  const noExcessiveScore = invertedScale(excessiveIterationRate * 150);
+  const overflowRate = (metrics.contextFillExceeded90Count ?? 0) / totalSessions;
+  const noOverflowScore = invertedScale(overflowRate * 120);
+  const bareRetryRate = (friction?.bareRetryAfterErrorCount ?? 0) / totalUtterances;
+  const noRetryScore = invertedScale(bareRetryRate * 300);
+  const frustrationRate = (friction?.frustrationExpressionCount ?? 0) / totalUtterances;
+  const noFrustrationScore = invertedScale(frustrationRate * 400);
+  const toolFailureRate = (friction?.toolFailureCount ?? 0) / Math.max(metrics.totalMessages, 1);
+  const noToolFailureScore = invertedScale(toolFailureRate * 200);
+  const mediumSessions = metrics.sessionHints?.mediumSessions ?? 0;
+  const focusBonus = mediumSessions / totalSessions * 15;
+  return clampScore(noExcessiveScore * 0.25 + noOverflowScore * 0.2 + noRetryScore * 0.2 + noFrustrationScore * 0.15 + noToolFailureScore * 0.1 + focusBonus + 10);
+}
 function computeDeterministicScores(phase1Output) {
   const metrics = phase1Output.sessionMetrics;
   return {
-    contextEfficiency: scoreContextEfficiency(metrics),
-    sessionOutcome: scoreSessionOutcome(metrics),
-    thinkingQuality: scoreThinkingQuality(metrics),
-    learningBehavior: scoreLearningBehavior(metrics),
-    communicationPatterns: scoreCommunicationPatterns(metrics, phase1Output),
+    aiPartnership: scoreAiPartnership(metrics),
+    sessionCraft: scoreSessionCraft(metrics),
+    toolMastery: scoreCommunicationPatterns(metrics, phase1Output),
+    skillResilience: scoreSkillResilience(metrics),
+    sessionMastery: scoreSessionMastery(metrics),
     controlScore: scoreControl(metrics)
   };
 }
@@ -29120,8 +29277,8 @@ function computeAffinities(scores, metrics, trendDensity) {
   const slashCmds = metrics.slashCommandCounts ?? {};
   const planCount = (slashCmds["plan"] ?? 0) + (slashCmds["review"] ?? 0);
   const planBonus = planCount > 0 ? Math.min(planCount * 8, 30) : 0;
-  const architectAffinity = scores.thinkingQuality * 0.5 + scores.controlScore * 0.3 + planBonus;
-  const analystAffinity = scores.thinkingQuality * 0.3 + scores.learningBehavior * 0.4 + scores.sessionOutcome * 0.2 + (metrics.questionRatio > 0.2 ? 10 : 0);
+  const architectAffinity = scores.aiPartnership * 0.5 + scores.controlScore * 0.3 + planBonus;
+  const analystAffinity = scores.aiPartnership * 0.3 + scores.sessionCraft * 0.4 + scores.sessionMastery * 0.2 + (metrics.questionRatio > 0.2 ? 10 : 0);
   const uniqueCommands = Object.keys(slashCmds).length;
   const totalCommands = Object.values(slashCmds).reduce((sum, c) => sum + c, 0);
   const orchestrationCmds = (slashCmds["sisyphus"] ?? 0) + (slashCmds["orchestrator"] ?? 0) + (slashCmds["ultrawork"] ?? 0) + (slashCmds["ralph-loop"] ?? 0);
@@ -29131,9 +29288,9 @@ function computeAffinities(scores, metrics, trendDensity) {
   const conductorAffinity = commandDiversityScore + commandVolumeScore + orchestrationBonus;
   const avgLen = metrics.avgDeveloperMessageLength;
   const concisenessScore = avgLen < 200 ? 40 : avgLen < 400 ? 25 : 10;
-  const speedrunnerAffinity = scores.contextEfficiency * 0.5 + concisenessScore + (scores.sessionOutcome > 70 ? 15 : 0);
+  const speedrunnerAffinity = scores.sessionCraft * 0.5 + concisenessScore + (scores.sessionMastery > 70 ? 15 : 0);
   const trendKeywordScore = trendDensity > 3 ? Math.min(trendDensity * 15, 60) : trendDensity * 5;
-  const learningCuriosityBonus = scores.learningBehavior > 70 ? 15 : 0;
+  const learningCuriosityBonus = scores.sessionCraft > 70 ? 15 : 0;
   const trendsetterAffinity = trendKeywordScore + learningCuriosityBonus;
   return {
     architect: architectAffinity,
@@ -29581,4 +29738,4 @@ export {
   clearAnalysisPending,
   shouldResumeRunningAnalysis
 };
-//# sourceMappingURL=chunk-72GWNTBD.js.map
+//# sourceMappingURL=chunk-SUEN2LKX.js.map

…ckages/plugin/dist/chunk-NH4BKYT6.js.map …ckages/plugin/dist/chunk-QFCYET5Y.js.mappackages/plugin/dist/chunk-NH4BKYT6.js.map renamed to packages/plugin/dist/chunk-QFCYET5Y.js.map packages/plugin/dist/chunk-NH4BKYT6.js.map renamed to packages/plugin/dist/chunk-QFCYET5Y.js.map

@@ -2,7 +2,7 @@
 import {
   readPrefs,
   writePrefs
-} from "../chunk-NH4BKYT6.js";
+} from "../chunk-QFCYET5Y.js";
 import {
   $ZodObject,
   $ZodType,
@@ -58,7 +58,7 @@ import {
   touchAnalysisHeartbeat,
   union,
   unknown
-} from "../chunk-72GWNTBD.js";
+} from "../chunk-SUEN2LKX.js";
 import {
   __commonJS,
   __require,

packages/plugin/dist/chunk-72GWNTBD.js packages/plugin/dist/chunk-SUEN2LKX.jspackages/plugin/dist/chunk-72GWNTBD.js renamed to packages/plugin/dist/chunk-SUEN2LKX.js

@@ -35,7 +35,9 @@ Each analysis skill runs as an **isolated Agent** rather than inline in this ses
 
 ## Agent Dispatch
 
-When `get_run_progress` returns a `nextStep` with a non-null `skill`, dispatch it as an Agent using the Agent tool:
+**CRITICAL**: Use the built-in `Agent` tool from your tool list to dispatch skills. Do NOT use Bash to run `claude` CLI commands (e.g., `claude --model haiku -p "..."`). The Agent tool is a first-class tool like `Read` or `Bash` -- invoke it directly with the parameters listed below.
+
+When `get_run_progress` returns a `nextStep` with a non-null `skill`, dispatch it as an Agent using the built-in `Agent` tool:
 
 - **prompt**: `"You are executing a BetterPrompt analysis stage. Read the skill instructions at [PLUGIN_PATH]/skills/[SKILL_NAME]/SKILL.md and follow them exactly. You have access to BetterPrompt MCP tools (get_prompt_context, get_stage_output, save_stage_output, save_domain_results, etc.). Execute the complete skill workflow. When finished, report whether the stage completed successfully or failed (include the error message if failed). Do not delegate to other agents or tasks."`
 - **model**: Use the model from the Model Tiering table above

…ckages/plugin/dist/chunk-72GWNTBD.js.map …ckages/plugin/dist/chunk-SUEN2LKX.js.mappackages/plugin/dist/chunk-72GWNTBD.js.map renamed to packages/plugin/dist/chunk-SUEN2LKX.js.map packages/plugin/dist/chunk-72GWNTBD.js.map renamed to packages/plugin/dist/chunk-SUEN2LKX.js.map

@@ -136,7 +136,12 @@ This is a one-time ask, tracked via `starAsked` in `~/.betterprompt/prefs.json`.
    > It helps others discover it: https://github.com/onlycastle/BetterPrompt"
    >
    > **"Sure, I'll star it"** / **"Skip"**
-3. Regardless of choice, write `starAsked: true` to prefs.
+3. If the user chose **"Sure, I'll star it"**, open the repo page in their browser:
+   ```bash
+   open "https://github.com/onlycastle/BetterPrompt"
+   ```
+   (Use `open` on macOS, `xdg-open` on Linux, `start` on Windows.)
+4. Regardless of choice, write `starAsked: true` to prefs.
    Prefer `save_user_prefs` instead of writing JSON manually.
 
 ### Step 6: Quick Reference + First Action
@@ -162,7 +167,7 @@ Then ask the user what they would like to do next.
 If the user already said to continue working or to run analysis now, follow that instruction directly and do not call `AskUserQuestion`.
 
 Use `AskUserQuestion` with these options:
-- **"Run bp analyze now"** (Recommended) — dispatch `bp-analyze` as an **Agent** (not as an inline skill) so it starts with a clean context. Use the Agent tool with:
+- **"Run bp analyze now"** (Recommended) — dispatch `bp-analyze` as an **Agent** (not as an inline skill) so it starts with a clean context. Use Claude Code's built-in `Agent` tool (do NOT use Bash to run `claude` CLI):
     ```
     model: sonnet
     description: "bp: analyze"

packages/plugin/dist/hooks/post-session-handler.js

@@ -39,6 +39,11 @@ describe('bp-analyze agent dispatch contract', () => {
     expect(content).toContain('must NOT internally spawn additional Agents or Tasks');
   });
 
+  it('explicitly warns against using Bash for agent dispatch', () => {
+    expect(content).toContain('Do NOT use Bash');
+    expect(content).toContain('built-in `Agent` tool');
+  });
+
   it('does not contain old single-session constraint language', () => {
     expect(content).not.toContain('single-session and deterministic');
     expect(content).not.toContain('Do NOT switch to `Agent`, `Task`, or any delegation mechanism');