codeharness 0.10.0 → 0.12.0

package/dist/index.js

@@ -1348,7 +1348,7 @@ function importStoriesToBeads(stories, opts, beadsFns) {
 }
 
 // src/commands/init.ts
-var HARNESS_VERSION = true ? "0.10.0" : "0.0.0-dev";
+var HARNESS_VERSION = true ? "0.12.0" : "0.0.0-dev";
 function getStackLabel(stack) {
   if (stack === "nodejs") return "Node.js (package.json)";
   if (stack === "python") return "Python";
@@ -2405,10 +2405,13 @@ function buildSpawnArgs(opts) {
   if (opts.live) {
     args.push("--live");
   }
+  if (opts.reset) {
+    args.push("--reset");
+  }
   return args;
 }
 function registerRunCommand(program) {
-  program.command("run").description("Execute the autonomous coding loop").option("--max-iterations <n>", "Maximum loop iterations", "50").option("--timeout <seconds>", "Total loop timeout in seconds", "14400").option("--iteration-timeout <minutes>", "Per-iteration timeout in minutes", "30").option("--live", "Show live output streaming", false).option("--calls <n>", "Max API calls per hour", "100").option("--max-story-retries <n>", "Max retries per story before flagging", "3").action(async (options, cmd) => {
+  program.command("run").description("Execute the autonomous coding loop").option("--max-iterations <n>", "Maximum loop iterations", "50").option("--timeout <seconds>", "Total loop timeout in seconds", "14400").option("--iteration-timeout <minutes>", "Per-iteration timeout in minutes", "30").option("--live", "Show live output streaming", false).option("--calls <n>", "Max API calls per hour", "100").option("--max-story-retries <n>", "Max retries per story before flagging", "3").option("--reset", "Clear retry counters, flagged stories, and circuit breaker before starting", false).action(async (options, cmd) => {
     const globalOpts = cmd.optsWithGlobals();
     const isJson = !!globalOpts.json;
     const outputOpts = { json: isJson };
@@ -2477,7 +2480,8 @@ function registerRunCommand(program) {
       iterationTimeout,
       calls,
       live: options.live,
-      maxStoryRetries
+      maxStoryRetries,
+      reset: options.reset
     });
     const env = { ...process.env };
     if (isJson) {
@@ -2592,6 +2596,32 @@ var INTEGRATION_KEYWORDS = [
   "integration test",
   "manual verification"
 ];
+var DOCKER_SESSION_KEYWORDS = [
+  "agent tool",
+  "subagent",
+  "via agent",
+  "invoke",
+  "/create-story",
+  "/bmad-dev-story",
+  "/bmad-code-review",
+  "/harness-run",
+  "/retrospective",
+  "sprint execution",
+  "fresh context",
+  "spawns",
+  "code review workflow",
+  "dev-story workflow",
+  "automatically proceeds",
+  "retries the current story",
+  "halts with status",
+  "prints summary"
+];
+var ESCALATE_KEYWORDS = [
+  "physical hardware",
+  "manual human",
+  "visual inspection by human",
+  "paid external service"
+];
 function classifyVerifiability(description) {
   const lower = description.toLowerCase();
   for (const kw of INTEGRATION_KEYWORDS) {
@@ -2599,6 +2629,19 @@ function classifyVerifiability(description) {
   }
   return "cli-verifiable";
 }
+function classifyStrategy(description) {
+  const lower = description.toLowerCase();
+  for (const kw of ESCALATE_KEYWORDS) {
+    if (lower.includes(kw)) return "escalate";
+  }
+  for (const kw of DOCKER_SESSION_KEYWORDS) {
+    if (lower.includes(kw)) return "docker-session";
+  }
+  for (const kw of INTEGRATION_KEYWORDS) {
+    if (lower.includes(kw)) return "docker-session";
+  }
+  return "cli-direct";
+}
 var VERIFICATION_TAG_PATTERN = /<!--\s*verification:\s*(cli-verifiable|integration-required)\s*-->/;
 function parseVerificationTag(text) {
   const match = VERIFICATION_TAG_PATTERN.exec(text);
@@ -2655,11 +2698,13 @@ function parseStoryACs(storyFilePath) {
       if (description) {
         const tag = parseVerificationTag(description);
         const verifiability = tag ?? classifyVerifiability(description);
+        const strategy = classifyStrategy(description);
         acs.push({
           id: currentId,
           description,
           type: classifyAC(description),
-          verifiability
+          verifiability,
+          strategy
         });
       } else {
         warn(`Skipping malformed AC #${currentId}: empty description`);
@@ -3294,27 +3339,52 @@ function validateProofQuality(proofPath) {
     return { verified: 0, pending: 0, escalated: 0, total: 0, passed: false };
   }
   const content = readFileSync10(proofPath, "utf-8");
-  const acHeaderPattern = /^## AC \d+:/gm;
+  const acHeaderPattern = /^## AC ?(\d+):/gm;
   const matches = [...content.matchAll(acHeaderPattern)];
-  if (matches.length === 0) {
-    return { verified: 0, pending: 0, escalated: 0, total: 0, passed: false };
-  }
   let verified = 0;
   let pending = 0;
   let escalated = 0;
-  for (let i = 0; i < matches.length; i++) {
-    const start = matches[i].index;
-    const end = i + 1 < matches.length ? matches[i + 1].index : content.length;
-    const section = content.slice(start, end);
-    if (section.includes("[ESCALATE]")) {
-      escalated++;
-      continue;
+  if (matches.length > 0) {
+    for (let i = 0; i < matches.length; i++) {
+      const start = matches[i].index;
+      const end = i + 1 < matches.length ? matches[i + 1].index : content.length;
+      const section = content.slice(start, end);
+      if (section.includes("[ESCALATE]")) {
+        escalated++;
+        continue;
+      }
+      const hasEvidence = section.includes("<!-- /showboat exec -->") || section.includes("<!-- showboat image:") || /```(?:bash|shell)\n[\s\S]*?```\n+```output\n/m.test(section);
+      if (hasEvidence) {
+        verified++;
+      } else {
+        pending++;
+      }
     }
-    const hasEvidence = section.includes("<!-- /showboat exec -->") || section.includes("<!-- showboat image:") || /```(?:bash|shell)\n[\s\S]*?```\n+```output\n/m.test(section);
-    if (hasEvidence) {
-      verified++;
-    } else {
-      pending++;
+  } else {
+    const inlineAcPattern = /--- AC ?(\d+):/g;
+    const inlineMatches = [...content.matchAll(inlineAcPattern)];
+    const acNumbers = new Set(inlineMatches.map((m) => m[1]));
+    if (acNumbers.size === 0) {
+      return { verified: 0, pending: 0, escalated: 0, total: 0, passed: false };
+    }
+    for (const acNum of acNumbers) {
+      const acPattern = new RegExp(`--- AC ?${acNum}:`, "g");
+      const acIdx = content.search(acPattern);
+      if (acIdx === -1) {
+        pending++;
+        continue;
+      }
+      const nextAcPattern = new RegExp(`--- AC ?(?!${acNum})\\d+:`, "g");
+      nextAcPattern.lastIndex = acIdx + 1;
+      const nextMatch = nextAcPattern.exec(content);
+      const section = content.slice(acIdx, nextMatch ? nextMatch.index : content.length);
+      if (section.includes("[ESCALATE]")) {
+        escalated++;
+      } else if (/```output\n/m.test(section)) {
+        verified++;
+      } else {
+        pending++;
+      }
     }
   }
   const total = verified + pending + escalated;
@@ -6783,7 +6853,7 @@ function registerGithubImportCommand(program) {
 }
 
 // src/index.ts
-var VERSION = true ? "0.10.0" : "0.0.0-dev";
+var VERSION = true ? "0.12.0" : "0.0.0-dev";
 function createProgram() {
   const program = new Command();
   program.name("codeharness").description("Makes autonomous coding agents produce software that actually works").version(VERSION).option("--json", "Output in machine-readable JSON format");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeharness",
-  "version": "0.10.0",
+  "version": "0.12.0",
   "type": "module",
   "description": "CLI for codeharness — makes autonomous coding agents produce software that actually works",
   "bin": {

package/ralph/ralph.sh

@@ -50,6 +50,9 @@ CLAUDE_OUTPUT_FORMAT="${CLAUDE_OUTPUT_FORMAT:-json}"
 CLAUDE_ALLOWED_TOOLS="${CLAUDE_ALLOWED_TOOLS:-}"
 CLAUDE_USE_CONTINUE="${CLAUDE_USE_CONTINUE:-false}"  # Fresh context per iteration by default
 
+# Reset retry state on start
+RESET_RETRIES=false
+
 # Live output
 LIVE_OUTPUT=false
 
@@ -722,6 +725,7 @@ Options:
     --prompt FILE             Prompt file for each iteration
     --progress FILE           Progress file (tasks JSON)
     --live                    Show live output streaming
+    --reset                   Clear retry counters, flagged stories, and circuit breaker before starting
     --reset-circuit           Reset circuit breaker and exit
     --status                  Show current status and exit
 
@@ -805,7 +809,20 @@ main() {
         fi
     fi
 
-    # Preserve retry state across restarts (Task 5.3)
+    # Reset retry state if --reset flag was passed
+    if [[ "$RESET_RETRIES" == "true" ]]; then
+        if [[ -f "$STORY_RETRY_FILE" ]]; then
+            rm -f "$STORY_RETRY_FILE"
+            log_status "INFO" "Cleared story retry counters"
+        fi
+        if [[ -f "$FLAGGED_STORIES_FILE" ]]; then
+            rm -f "$FLAGGED_STORIES_FILE"
+            log_status "INFO" "Cleared flagged stories"
+        fi
+        reset_circuit_breaker "Reset via --reset flag"
+        log_status "INFO" "Circuit breaker reset to CLOSED"
+    fi
+
     # .story_retries and .flagged_stories are file-based — they persist automatically
 
     log_status "SUCCESS" "Ralph loop starting"
@@ -1056,6 +1073,10 @@ while [[ $# -gt 0 ]]; do
             LIVE_OUTPUT=true
             shift
             ;;
+        --reset)
+            RESET_RETRIES=true
+            shift
+            ;;
         --reset-circuit)
             # Derive state paths so circuit breaker uses the correct directory
             HARNESS_STATE_DIR="$(pwd)/.claude"