pan-wizard 3.5.2 → 3.8.0

package/hooks/dist/pan-trace-logger.js

@@ -80,6 +80,65 @@ function extractNumber(obj, key) {
   return typeof v === 'number' ? v : 0;
 }
 
+/**
+ * P-1805 (v3.7.8): extract usage totals by reading the SubagentStop transcript.
+ * The hook payload from Claude Code in headless mode does NOT include
+ * `data.usage` — only `transcript_path`. This function reads the transcript
+ * JSONL file and sums the `usage` fields across all assistant messages
+ * belonging to the just-completed subagent (its session_id is in `data.session_id`).
+ *
+ * Returns the SUM of input/output/cache tokens for the subagent's whole
+ * conversation. Agents that finish in one Claude API call will have one
+ * usage record; agents with multi-turn tool use will have several.
+ *
+ * Returns `{input_tokens, output_tokens, cache_read_input_tokens, cache_creation_input_tokens}`
+ * with all zeros if the transcript is unreadable / missing — same fallback
+ * shape as the original behavior, so callers don't break.
+ *
+ * @param {string} transcriptPath - Absolute path to the transcript JSONL file
+ * @param {string} [sessionId] - Optional subagent session_id to filter on
+ * @returns {Object} usage totals object
+ */
+function readUsageFromTranscript(transcriptPath, sessionId) {
+  const totals = {
+    input_tokens: 0,
+    output_tokens: 0,
+    cache_read_input_tokens: 0,
+    cache_creation_input_tokens: 0,
+  };
+  if (!transcriptPath || typeof transcriptPath !== 'string') return totals;
+  let raw;
+  try {
+    raw = fs.readFileSync(transcriptPath, 'utf-8');
+  } catch {
+    return totals;
+  }
+  for (const line of raw.split('\n')) {
+    if (!line) continue;
+    let entry;
+    try {
+      entry = JSON.parse(line);
+    } catch {
+      continue;
+    }
+    // Filter to entries from this subagent if a session_id is provided.
+    // The transcript may include parent + child traffic; session_id discriminates.
+    if (sessionId && entry.session_id && entry.session_id !== sessionId) continue;
+    // Usage typically lives on assistant message records.
+    const usage = entry.usage
+      || entry.message?.usage
+      || entry.response?.usage
+      || (entry.type === 'assistant' && entry.message?.usage)
+      || null;
+    if (!usage || typeof usage !== 'object') continue;
+    totals.input_tokens += extractNumber(usage, 'input_tokens');
+    totals.output_tokens += extractNumber(usage, 'output_tokens');
+    totals.cache_read_input_tokens += extractNumber(usage, 'cache_read_input_tokens');
+    totals.cache_creation_input_tokens += extractNumber(usage, 'cache_creation_input_tokens');
+  }
+  return totals;
+}
+
 /**
  * Build trace event(s) from a SubagentStop payload.
  * Pure function — no side effects.
@@ -93,9 +152,19 @@ function buildTraceEvents(data, sessionId) {
 
   const ts = new Date().toISOString();
   const agent = data.agent_type || data.subagent_type || 'unknown';
-  const inputTokens = extractNumber(data.usage, 'input_tokens');
-  const outputTokens = extractNumber(data.usage, 'output_tokens');
-  const cacheRead = extractNumber(data.usage, 'cache_read_input_tokens');
+
+  // P-1805: prefer usage from data.usage when present (interactive Claude Code path).
+  // Fall back to reading the transcript file (headless `claude -p` path — usage
+  // not in payload but discoverable via transcript_path).
+  let inputTokens = extractNumber(data.usage, 'input_tokens');
+  let outputTokens = extractNumber(data.usage, 'output_tokens');
+  let cacheRead = extractNumber(data.usage, 'cache_read_input_tokens');
+  if ((inputTokens + outputTokens + cacheRead) === 0 && data.transcript_path) {
+    const fromTranscript = readUsageFromTranscript(data.transcript_path, data.session_id);
+    inputTokens = fromTranscript.input_tokens;
+    outputTokens = fromTranscript.output_tokens;
+    cacheRead = fromTranscript.cache_read_input_tokens;
+  }
   const totalTokens = inputTokens + outputTokens;
 
   const events = [];

package/package.json

@@ -1,66 +1,67 @@
-{
-  "name": "pan-wizard",
-  "version": "3.5.2",
-  "description": "A lightweight workflow automation and context engineering system for Claude Code, OpenCode, Gemini CLI, Codex, and Copilot CLI.",
-  "bin": {
-    "pan-wizard": "bin/install.js"
-  },
-  "files": [
-    "bin",
-    "commands",
-    "pan-wizard-core",
-    "agents",
-    "hooks/dist",
-    "scripts",
-    "assets"
-  ],
-  "keywords": [
-    "claude",
-    "claude-code",
-    "ai",
-    "workflow-automation",
-    "context-engineering",
-    "project-automation",
-    "gemini",
-    "gemini-cli",
-    "codex",
-    "codex-cli",
-    "copilot",
-    "copilot-cli",
-    "github-copilot"
-  ],
-  "author": "PAN Wizard Contributors",
-  "contributors": [
-    {
-      "name": "oharms",
-      "url": "https://github.com/oharms"
-    }
-  ],
-  "license": "MIT",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/oharms/PanWizard.git"
-  },
-  "homepage": "https://github.com/oharms/PanWizard#readme",
-  "bugs": {
-    "url": "https://github.com/oharms/PanWizard/issues"
-  },
-  "engines": {
-    "node": ">=16.7.0"
-  },
-  "devDependencies": {
-    "@playwright/test": "^1.58.2",
-    "@vscode/test-electron": "^2.5.2",
-    "esbuild": "^0.24.0"
-  },
-  "scripts": {
-    "build:hooks": "node scripts/build-hooks.js",
-    "prepublishOnly": "npm run build:hooks",
-    "test": "node --test tests/*.test.cjs",
-    "test:scenarios": "node --test tests/scenarios/*.test.cjs",
-    "test:all": "node --test tests/*.test.cjs tests/scenarios/*.test.cjs",
-    "test:e2e": "node --test tests/scenarios/*.test.cjs",
-    "test:vscode": "npx playwright test --config tests/e2e/playwright.config.mjs",
-    "test:watch": "node --test --watch tests/*.test.cjs"
-  }
-}
+{
+  "name": "pan-wizard",
+  "version": "3.8.0",
+  "description": "A lightweight workflow automation and context engineering system for Claude Code, OpenCode, Gemini CLI, Codex, and Copilot CLI.",
+  "bin": {
+    "pan-wizard": "bin/install.js"
+  },
+  "files": [
+    "bin",
+    "commands",
+    "pan-wizard-core",
+    "agents",
+    "hooks/dist",
+    "scripts",
+    "assets"
+  ],
+  "keywords": [
+    "claude",
+    "claude-code",
+    "ai",
+    "workflow-automation",
+    "context-engineering",
+    "project-automation",
+    "gemini",
+    "gemini-cli",
+    "codex",
+    "codex-cli",
+    "copilot",
+    "copilot-cli",
+    "github-copilot"
+  ],
+  "author": "PAN Wizard Contributors",
+  "contributors": [
+    {
+      "name": "oharms",
+      "url": "https://github.com/oharms"
+    }
+  ],
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/oharms/PanWizard.git"
+  },
+  "homepage": "https://github.com/oharms/PanWizard#readme",
+  "bugs": {
+    "url": "https://github.com/oharms/PanWizard/issues"
+  },
+  "engines": {
+    "node": ">=16.7.0"
+  },
+  "devDependencies": {
+    "@playwright/test": "^1.58.2",
+    "@vscode/test-electron": "^2.5.2",
+    "esbuild": "^0.28.0"
+  },
+  "scripts": {
+    "build:hooks": "node scripts/build-hooks.js",
+    "release:check": "node scripts/release-check.js",
+    "prepublishOnly": "node scripts/release-check.js",
+    "test": "node --test tests/*.test.cjs",
+    "test:scenarios": "node --test tests/scenarios/*.test.cjs",
+    "test:all": "node --test tests/*.test.cjs tests/scenarios/*.test.cjs",
+    "test:e2e": "node --test tests/scenarios/*.test.cjs",
+    "test:vscode": "npx playwright test --config tests/e2e/playwright.config.mjs",
+    "test:watch": "node --test --watch tests/*.test.cjs"
+  }
+}

package/pan-wizard-core/bin/lib/codebase.cjs

@@ -5,6 +5,8 @@
  * Supports JS/TS (v0), with extensible language registry for Python/Go/Rust/Java/C# (v1).
  */
 
+// @pan: ADR-0021
+
 const fs = require('fs');
 const path = require('path');
 const { CODEBASE_DIR, DRIFT_MAX_FILE_SIZE } = require('./constants.cjs');

package/pan-wizard-core/bin/lib/commands.cjs

@@ -379,6 +379,11 @@ function runCommitSafetyChecks(cwd, config, force) {
 function cmdCommit(cwd, message, files, raw, amend, opts) {
   const commitType = opts && opts.type;
   const force = opts && opts.force;
+  // P-EXP-001 follow-up (v3.7.10): when failOnError is true, commit_failed
+  // exits non-zero so callers (especially autonomous loops) detect the silent-
+  // failure case where git refused (e.g. missing identity) and the loop would
+  // otherwise keep going thinking the artifact landed.
+  const failOnError = opts && opts.failOnError;
 
   if (!isGitRepo(cwd)) {
     output({ committed: false, hash: null, reason: 'not_a_git_repo', hint: 'Run git init to initialize a repository' }, raw, 'not a git repo');
@@ -425,6 +430,9 @@ function cmdCommit(cwd, message, files, raw, amend, opts) {
       output({ committed: false, hash: null, reason: 'nothing_to_commit' }, raw, 'nothing');
       return;
     }
+    if (failOnError) {
+      error('commit_failed: ' + (commitResult.stderr || 'unknown git error').trim());
+    }
     output({ committed: false, hash: null, reason: 'commit_failed', error: commitResult.stderr }, raw, 'failed');
     return;
   }

package/pan-wizard-core/bin/lib/config.cjs

@@ -52,6 +52,7 @@ function buildConfigDefaults(hasBraveSearch, userDefaults) {
       plan_check: true,
       verifier: true,
       nyquist_validation: false,
+      phase_record_compact: false,
     },
     parallelization: true,
     brave_search: hasBraveSearch,
@@ -422,17 +423,27 @@ function cmdStandardsRecommend(cwd, raw) {
   if (/\b(enterprise|togaf|architecture\s*governance|compliance)\b/.test(lower)) detectedTypes.push('enterprise');
   if (/\b(cli|command.line|terminal|shell|argv)\b/.test(lower)) detectedTypes.push('cli');
 
-  if (detectedTypes.length === 0) detectedTypes.push('general');
+  const explicitlyDetected = detectedTypes.length > 0;
+  if (!explicitlyDetected) detectedTypes.push('general');
 
   const seen = new Set();
   const recommendations = [];
   for (const type of detectedTypes) {
     const recs = STANDARDS_RECOMMENDATIONS[type] || [];
+    let perTypeIndex = 0;
     for (const id of recs) {
       if (seen.has(id)) continue;
       seen.add(id);
       const s = STANDARDS_CATALOG[id];
-      recommendations.push({ id, name: s.name, reason: type + ' project detected', priority: recommendations.length < 3 ? 'high' : 'medium' });
+      const isHigh = explicitlyDetected && type !== 'general' && perTypeIndex === 0;
+      recommendations.push({
+        id,
+        name: s.name,
+        reason: type + ' project detected',
+        priority: isHigh ? 'high' : 'medium',
+        source_type: type,
+      });
+      perTypeIndex += 1;
     }
   }
 

package/pan-wizard-core/bin/lib/context-budget.cjs

@@ -21,6 +21,51 @@ function estimateTokens(text) {
   return Math.ceil(text.length / CHARS_PER_TOKEN);
 }
 
+/**
+ * Estimate a relevance signal for per-phase markdown content (P-RES-002).
+ *
+ * Per Chroma's "Context Rot" research (Hong & Huber, July 2025): a single
+ * semantically-similar-but-irrelevant distractor degrades performance even
+ * at modest context sizes. Distractor density matters more than token count.
+ *
+ * Computing TRUE topic-relevance requires embeddings or keyword analysis
+ * we can't do cheaply at zero deps. This v0 heuristic reports a simpler
+ * signal: structure-vs-content ratio. Markdown files heavy on headers,
+ * separators, empty bullet lists, and placeholder text are LESS dense in
+ * actual signal than files of equal length with concrete prose. The ratio
+ * isn't true distractor density but is correlated with it for the
+ * "thin/template-only context" failure mode.
+ *
+ * Returns ratio in [0, 1] where 1 = all content lines, 0 = all structure.
+ * Returns null if not enough lines to compute meaningfully.
+ *
+ * @param {string} text
+ * @returns {number|null}
+ */
+function estimateRelevanceRatio(text) {
+  if (!text) return null;
+  const lines = text.split(/\r?\n/);
+  if (lines.length < 5) return null;
+  let contentLines = 0;
+  let totalLines = 0;
+  for (const raw of lines) {
+    const line = raw.trim();
+    if (!line) continue;                              // skip blank
+    totalLines++;
+    if (/^#{1,6}\s/.test(line)) continue;             // skip header
+    if (/^[-*_]{3,}$/.test(line)) continue;           // skip separator
+    if (/^[-*]\s*$/.test(line)) continue;             // skip empty bullet
+    if (/^[-*]\s*\[\s*[\]x_]\s*\]\s*$/.test(line)) continue; // empty checkbox
+    if (/^\|\s*-+\s*\|/.test(line)) continue;         // table separator row
+    if (/^>\s*$/.test(line)) continue;                // empty blockquote
+    if (line.length < 10) continue;                   // very short — likely scaffolding
+    if (/^(TODO|TBD|FIXME|placeholder|todo|tbd|fixme|placeholder|coming soon)\b/i.test(line)) continue;
+    contentLines++;
+  }
+  if (totalLines === 0) return null;
+  return Math.round((contentLines / totalLines) * 1000) / 1000;
+}
+
 /**
  * Compute context budget for the current project state.
  * @param {string} cwd - Project root directory
@@ -101,6 +146,32 @@ function cmdContextBudget(cwd, raw) {
     recommendation = `Within budget. ~${additionalPlans} more plans could fit before degradation.`;
   }
 
+  // P-RES-002 signal: structure-vs-content ratio for per-phase markdown.
+  // High structure (lots of empty bullets, headers, placeholders) suggests
+  // the per-phase context is thin/templatey rather than substantive — a
+  // proxy for "context that's wasting tokens on filler."
+  let relevanceSignal = null;
+  if (phaseDir) {
+    const fullPhasePath = path.join(cwd, phaseDir);
+    const samples = [];
+    for (const fname of ['research.md', 'context.md']) {
+      const candidate = path.join(fullPhasePath, fname);
+      const content = safeReadFile(candidate);
+      if (content) {
+        const ratio = estimateRelevanceRatio(content);
+        if (ratio !== null) samples.push({ file: toPosix(path.join(phaseDir, fname)), ratio });
+      }
+    }
+    if (samples.length > 0) {
+      const avg = samples.reduce((a, s) => a + s.ratio, 0) / samples.length;
+      relevanceSignal = {
+        avg_ratio: Math.round(avg * 1000) / 1000,
+        samples,
+        note: 'P-RES-002 v0 heuristic — structure/content ratio. <0.4 suggests thin per-phase context (heavy on headers + empty buckets + placeholders).',
+      };
+    }
+  }
+
   // E-8: cache metrics — surface how much of the total context would be
   // served from prompt cache when Opus 4.7 cache_control is active.
   const { buildCachedContext } = require('./core.cjs');
@@ -141,6 +212,7 @@ function cmdContextBudget(cwd, raw) {
     contextWindow: CONTEXT_WINDOW,
     budgetUtilization: Math.round(utilization * 1000) / 1000,
     cache,
+    relevanceSignal,
     recommendation,
   };
 
@@ -174,4 +246,5 @@ function cmdContextBudget(cwd, raw) {
 module.exports = {
   cmdContextBudget,
   estimateTokens,
+  estimateRelevanceRatio,
 };

package/pan-wizard-core/bin/lib/core.cjs

@@ -48,6 +48,7 @@ const COST_MULTIPLIERS = { reasoning: 15, mid: 3, fast: 1 };
 // ─── Model Profile Table ─────────────────────────────────────────────────────
 
 const MODEL_PROFILES = {
+  // Original planning/execution agents (pre-v3.0)
   'pan-planner':              { quality: 'reasoning', balanced: 'reasoning', budget: 'mid' },
   'pan-roadmapper':           { quality: 'reasoning', balanced: 'mid',      budget: 'mid' },
   'pan-executor':             { quality: 'reasoning', balanced: 'mid',      budget: 'mid' },
@@ -60,6 +61,18 @@ const MODEL_PROFILES = {
   'pan-plan-checker':         { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
   'pan-integration-checker':  { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
   'pan-reviewer':             { quality: 'reasoning', balanced: 'fast',     budget: 'fast' },
+  // Spec B v2 agents (v3.0–v3.4) — added v3.7.5 to close MODEL_PROFILES drift
+  'pan-conductor':            { quality: 'reasoning', balanced: 'reasoning', budget: 'mid' },
+  'pan-counterfactual':       { quality: 'reasoning', balanced: 'mid',      budget: 'mid' },
+  'pan-hardener':             { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
+  'pan-meta-reviewer':        { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
+  'pan-knowledge':            { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
+  'pan-previewer':            { quality: 'reasoning', balanced: 'fast',     budget: 'fast' },
+  // v3.5 agents
+  'pan-optimizer':            { quality: 'reasoning', balanced: 'mid',      budget: 'fast' },
+  'pan-distiller':            { quality: 'reasoning', balanced: 'fast',     budget: 'fast' },
+  // v3.7.0 self-improvement loop — observation-only watchdog
+  'pan-experiment-runner':    { quality: 'reasoning', balanced: 'fast',     budget: 'fast' },
 };
 
 // ─── Output helpers ───────────────────────────────────────────────────────────