From b9799eef090f2592b3c973ff75c2a39c0de7f2e3 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:30:28 +0000
Subject: [PATCH 01/14] Initial plan


From 43091955ed7395e154ff593aa665aa2f55d91469 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:47:50 +0000
Subject: [PATCH 02/14] fix: improve Claude empty-startup diagnostics and retry
 handling

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs           | 45 ++++++++++++-
 actions/setup/js/claude_harness.test.cjs      | 27 ++++++++
 actions/setup/js/log_parser_bootstrap.cjs     | 65 ++++++++++++++++++-
 .../setup/js/log_parser_bootstrap.test.cjs    | 31 ++++++++-
 actions/setup/js/parse_claude_log.test.cjs    |  4 +-
 5 files changed, 166 insertions(+), 6 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 142eaaa3246..1b38d0dce4a 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -99,6 +99,35 @@ function resolveRetryConfig(env = process.env, logger = log) {
   return resolveSharedRetryConfig(env, logger);
 }
 
+/**
+ * Parse bounded startup retries for zero-output startup failures.
+ * Retries are always fresh runs (never --continue).
+ *
+ * @param {NodeJS.ProcessEnv} [env]
+ * @param {(message: string) => void} [logger]
+ * @returns {number}
+ */
+function resolveStartupRetryLimit(env = process.env, logger = log) {
+  const raw = env.GH_AW_CLAUDE_STARTUP_RETRIES;
+  if (raw == null || raw === "") {
+    return 1;
+  }
+  if (!/^[+-]?\d+$/.test(raw)) {
+    logger(`invalid GH_AW_CLAUDE_STARTUP_RETRIES='${raw}' (expected integer); using default startup retries=1`);
+    return 1;
+  }
+  const parsed = Number.parseInt(raw, 10);
+  if (!Number.isFinite(parsed)) {
+    logger(`invalid GH_AW_CLAUDE_STARTUP_RETRIES='${raw}' (not finite); using default startup retries=1`);
+    return 1;
+  }
+  const clamped = Math.min(Math.max(parsed, 0), 2);
+  if (clamped !== parsed) {
+    logger(`GH_AW_CLAUDE_STARTUP_RETRIES=${parsed} out of range; clamped to ${clamped}`);
+  }
+  return clamped;
+}
+
 /**
  * Determines if the collected output contains an Anthropic overload error.
  * @param {string} output - Collected stdout+stderr from the process
@@ -297,6 +326,7 @@ async function main() {
   const [, , command, ...args] = process.argv;
   const retryConfig = resolveRetryConfig(process.env, log);
   const { maxRetries, initialDelayMs, backoffMultiplier, maxDelayMs } = retryConfig;
+  const startupRetryLimit = resolveStartupRetryLimit(process.env, log);
 
   if (!command) {
     process.stderr.write("claude-harness: Usage: node claude_harness.cjs <command> [args...]\n");
@@ -347,6 +377,7 @@ async function main() {
   let lastExitCode = 1;
   let useContinueOnRetry = false;
   let continueDisabledPermanently = false;
+  let startupRetriesUsed = 0;
   const driverStartTime = Date.now();
   // Soft-timeout guard: polled at the top of the retry loop and after each backoff sleep.
   // It does not preempt a running attempt — if a single invocation runs past the soft
@@ -519,7 +550,18 @@ async function main() {
     if (attempt >= maxRetries) {
       log(`all ${maxRetries} retries exhausted — giving up (exitCode=${lastExitCode})`);
     } else {
-      log(`attempt ${attempt + 1}: no output produced — not retrying` + ` (possible causes: binary not found, permission denied, auth failure, or silent startup crash)`);
+      if (startupRetriesUsed < startupRetryLimit) {
+        startupRetriesUsed++;
+        useContinueOnRetry = false;
+        continueDisabledPermanently = true;
+        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, attempt ${attempt + 2}/${maxRetries + 1})`);
+        continue;
+      }
+      log(
+        `attempt ${attempt + 1}: no output produced — not retrying` +
+          ` (startup retry budget exhausted: ${startupRetriesUsed}/${startupRetryLimit};` +
+          ` possible causes: binary not found, permission denied, auth failure, or silent startup crash)`
+      );
     }
 
     break;
@@ -551,6 +593,7 @@ if (typeof module !== "undefined" && module.exports) {
     hasNoopInSafeOutputs,
     hasExpectedSafeOutputs,
     resolveRetryConfig,
+    resolveStartupRetryLimit,
   };
 }
 
diff --git a/actions/setup/js/claude_harness.test.cjs b/actions/setup/js/claude_harness.test.cjs
index 4d888d13808..c9d09dbf35d 100644
--- a/actions/setup/js/claude_harness.test.cjs
+++ b/actions/setup/js/claude_harness.test.cjs
@@ -21,6 +21,7 @@ const {
   extractDeniedCommands,
   buildMissingToolPermissionIssuePayload,
   resolveRetryConfig,
+  resolveStartupRetryLimit,
 } = require("./claude_harness.cjs");
 
 const agentTempDir = "/tmp/gh-aw/agent";
@@ -448,6 +449,24 @@ process.exit(0);
       expect(result.stderr).toContain("failure_reason=cancelled_or_timed_out");
     }, 30000);
 
+    it("retries one no-output startup failure as a fresh run by default", () => {
+      const stubScript = `
+const fs = require("fs");
+const callsPath = process.env.CLAUDE_HARNESS_STUB_CALLS;
+const args = process.argv.slice(2);
+const priorCalls = fs.existsSync(callsPath) ? fs.readFileSync(callsPath, "utf8").trim().split("\\n").filter(Boolean).length : 0;
+fs.appendFileSync(callsPath, JSON.stringify({ args }) + "\\n", "utf8");
+if (priorCalls === 0) process.exit(1);
+process.stdout.write("startup retry succeeded\\n");
+process.exit(0);
+`;
+      const { result, calls } = runHarnessWithStub({ stubScript });
+      expect(result.status, result.stderr).toBe(0);
+      expect(calls.length).toBe(2);
+      expect(calls[1].args.includes("--continue")).toBe(false);
+      expect(result.stderr).toContain("no output produced — retrying startup as fresh run");
+    });
+
     it("returns true for normal partial-execution retry", () => {
       const result = shouldRetryWithContinue({
         attempt: 0,
@@ -559,6 +578,14 @@ process.exit(0);
       const config2 = resolveRetryConfig({ GH_AW_HARNESS_INITIAL_DELAY_MS: "0x10" });
       expect(config2.initialDelayMs).toBe(5000);
     });
+
+    it("uses startup retry default and clamps overrides to [0..2]", () => {
+      expect(resolveStartupRetryLimit({})).toBe(1);
+      expect(resolveStartupRetryLimit({ GH_AW_CLAUDE_STARTUP_RETRIES: "2" })).toBe(2);
+      expect(resolveStartupRetryLimit({ GH_AW_CLAUDE_STARTUP_RETRIES: "-5" })).toBe(0);
+      expect(resolveStartupRetryLimit({ GH_AW_CLAUDE_STARTUP_RETRIES: "9" })).toBe(2);
+      expect(resolveStartupRetryLimit({ GH_AW_CLAUDE_STARTUP_RETRIES: "nope" })).toBe(1);
+    });
   });
 
   describe("noop pre-flight and retry guard", () => {
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 26d420596d0..c1032ef3970 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -4,6 +4,54 @@
 const { generatePlainTextSummary, generateCopilotCliStyleSummary, wrapAgentLogInSection, formatSafeOutputsPreview } = require("./log_parser_shared.cjs");
 const { getErrorMessage } = require("./error_helpers.cjs");
 const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
+const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
+const CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit|overloaded_error|"overloaded"|5\d\d/i;
+
+/**
+ * Build startup diagnostics for Claude failures with no structured entries.
+ * @param {string} rawContent
+ * @returns {{
+ *   exitCode: string,
+ *   inferenceAccessError: boolean,
+ *   aiCreditsRateLimitError: boolean,
+ *   summaryLine: string,
+ *   summaryMarkdown: string
+ * }}
+ */
+function buildClaudeStartupDiagnostics(rawContent) {
+  const content = typeof rawContent === "string" ? rawContent : "";
+  const lines = content
+    .split(/\r?\n/)
+    .map(line => line.trim())
+    .filter(Boolean);
+  const startupLines = lines.filter(line => line.includes("[claude-harness]") || /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i.test(line));
+  const tailLines = (startupLines.length > 0 ? startupLines : lines).slice(-8);
+  const tailText = tailLines.join("\n");
+
+  let exitCode = "unknown";
+  const doneMatches = [...content.matchAll(/done:\s*exitCode=(\d+)/g)];
+  if (doneMatches.length > 0) {
+    exitCode = doneMatches[doneMatches.length - 1][1];
+  } else {
+    const attemptMatches = [...content.matchAll(/attempt\s+\d+\s+failed:\s+exitCode=(\d+)/g)];
+    if (attemptMatches.length > 0) {
+      exitCode = attemptMatches[attemptMatches.length - 1][1];
+    }
+  }
+
+  const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
+  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN.test(content);
+  const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
+  const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
+
+  return {
+    exitCode,
+    inferenceAccessError,
+    aiCreditsRateLimitError,
+    summaryLine,
+    summaryMarkdown,
+  };
+}
 
 /**
  * Bootstrap helper for log parser entry points.
@@ -344,7 +392,22 @@ async function runLogParser(options) {
           `Claude produced no structured log entries, but agent completed with ${safeOutputEntriesCount} safe output ${safeOutputEntriesCount === 1 ? "entry" : "entries"} — treating as non-fatal post-completion infrastructure failure`
         );
       } else {
-        core.setFailed(`${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. This usually indicates a startup or configuration error before tool execution.`);
+        const diagnostics = buildClaudeStartupDiagnostics(content);
+        if (diagnostics.summaryMarkdown) {
+          await core.summary.addRaw(diagnostics.summaryMarkdown).write();
+        }
+
+        if (diagnostics.inferenceAccessError) {
+          core.setOutput("inference_access_error", "true");
+        }
+        if (diagnostics.aiCreditsRateLimitError) {
+          core.setOutput("ai_credits_rate_limit_error", "true");
+        }
+
+        const errorCode = diagnostics.inferenceAccessError || diagnostics.aiCreditsRateLimitError ? ERR_API : ERR_CONFIG;
+        const failureKind = diagnostics.inferenceAccessError || diagnostics.aiCreditsRateLimitError ? "transient inference availability signal detected" : "startup/configuration failure detected";
+
+        core.setFailed(`${errorCode}: Claude execution failed: no structured log entries were produced. ${diagnostics.summaryLine} ${failureKind}.`);
       }
     }
 
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index 19f9685a9f8..8eff4688e3c 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -71,7 +71,9 @@ describe("log_parser_bootstrap.cjs", () => {
             process.env.GH_AW_AGENT_OUTPUT = logFile;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
             await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
-            expect(mockCore.setFailed).toHaveBeenCalledWith(`${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. This usually indicates a startup or configuration error before tool execution.`);
+            expect(mockCore.setFailed).toHaveBeenCalledWith(
+              `${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=unknown). startup/configuration failure detected.`
+            );
           } finally {
             fs.unlinkSync(logFile);
             fs.rmdirSync(tmpDir);
@@ -237,7 +239,9 @@ describe("log_parser_bootstrap.cjs", () => {
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
             await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
-            expect(mockCore.setFailed).toHaveBeenCalledWith(`${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. This usually indicates a startup or configuration error before tool execution.`);
+            expect(mockCore.setFailed).toHaveBeenCalledWith(
+              `${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=unknown). startup/configuration failure detected.`
+            );
           } finally {
             fs.unlinkSync(logFile);
             fs.rmdirSync(tmpDir);
@@ -274,7 +278,28 @@ describe("log_parser_bootstrap.cjs", () => {
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: null });
             await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
-            expect(mockCore.setFailed).toHaveBeenCalledWith(`${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. This usually indicates a startup or configuration error before tool execution.`);
+            expect(mockCore.setFailed).toHaveBeenCalledWith(
+              `${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=unknown). startup/configuration failure detected.`
+            );
+          } finally {
+            fs.unlinkSync(logFile);
+            delete process.env.GH_AW_AGENT_OUTPUT;
+            fs.rmdirSync(tmpDir);
+          }
+        }),
+        it("should classify Claude empty-log startup rate-limit signatures as transient inference availability", async () => {
+          const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
+          const logFile = path.join(tmpDir, "test.log");
+          try {
+            fs.writeFileSync(logFile, `[claude-harness] attempt 1 failed: exitCode=1 isRateLimitError=true hasOutput=false\n[claude-harness] done: exitCode=1\nAPI Error: Request rejected (429)`);
+            process.env.GH_AW_AGENT_OUTPUT = logFile;
+            delete process.env.GH_AW_SAFE_OUTPUTS;
+            const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
+            await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
+            expect(mockCore.setFailed).toHaveBeenCalledWith(
+              `${ERR_API}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=1). transient inference availability signal detected.`
+            );
+            expect(mockCore.setOutput).toHaveBeenCalledWith("ai_credits_rate_limit_error", "true");
           } finally {
             fs.unlinkSync(logFile);
             delete process.env.GH_AW_AGENT_OUTPUT;
diff --git a/actions/setup/js/parse_claude_log.test.cjs b/actions/setup/js/parse_claude_log.test.cjs
index 09baa6d21bb..bd724a276c9 100644
--- a/actions/setup/js/parse_claude_log.test.cjs
+++ b/actions/setup/js/parse_claude_log.test.cjs
@@ -441,7 +441,9 @@ describe("parse_claude_log.cjs", () => {
 
     it("should fail when Claude log has no structured entries", async () => {
       await runScript("this is not structured Claude JSON output");
-      expect(mockCore.setFailed).toHaveBeenCalledWith(`${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. This usually indicates a startup or configuration error before tool execution.`);
+      expect(mockCore.setFailed).toHaveBeenCalledWith(
+        `${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=unknown). startup/configuration failure detected.`
+      );
     });
   });
 

From 0ccd4ab36f801ad240bb26657ca980a9778d597f Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:51:49 +0000
Subject: [PATCH 03/14] refine: tighten startup signal matching and retry
 behavior

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       | 1 -
 actions/setup/js/log_parser_bootstrap.cjs | 2 +-
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 1b38d0dce4a..f93648b9237 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -553,7 +553,6 @@ async function main() {
       if (startupRetriesUsed < startupRetryLimit) {
         startupRetriesUsed++;
         useContinueOnRetry = false;
-        continueDisabledPermanently = true;
         log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, attempt ${attempt + 2}/${maxRetries + 1})`);
         continue;
       }
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index c1032ef3970..497ec39288a 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -5,7 +5,7 @@ const { generatePlainTextSummary, generateCopilotCliStyleSummary, wrapAgentLogIn
 const { getErrorMessage } = require("./error_helpers.cjs");
 const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
-const CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit|overloaded_error|"overloaded"|5\d\d/i;
+const CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit|overloaded_error|"overloaded"|(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
 
 /**
  * Build startup diagnostics for Claude failures with no structured entries.

From c87d2807b89ed0f0b0ca32b27fb0e9b1a64c432a Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:55:46 +0000
Subject: [PATCH 04/14] chore: address review feedback on startup diagnostics

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       | 2 +-
 actions/setup/js/log_parser_bootstrap.cjs | 9 ++++++---
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index f93648b9237..e9b630b278d 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -553,7 +553,7 @@ async function main() {
       if (startupRetriesUsed < startupRetryLimit) {
         startupRetriesUsed++;
         useContinueOnRetry = false;
-        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, attempt ${attempt + 2}/${maxRetries + 1})`);
+        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${attempt + 2} of ${maxRetries + 1} total attempts)`);
         continue;
       }
       log(
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 497ec39288a..b9270f746c7 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -5,7 +5,10 @@ const { generatePlainTextSummary, generateCopilotCliStyleSummary, wrapAgentLogIn
 const { getErrorMessage } = require("./error_helpers.cjs");
 const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
-const CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit|overloaded_error|"overloaded"|(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
+const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
+const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
+const CLAUDE_HTTP_5XX_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
+const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i;
 
 /**
  * Build startup diagnostics for Claude failures with no structured entries.
@@ -24,7 +27,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
     .split(/\r?\n/)
     .map(line => line.trim())
     .filter(Boolean);
-  const startupLines = lines.filter(line => line.includes("[claude-harness]") || /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i.test(line));
+  const startupLines = lines.filter(line => line.includes("[claude-harness]") || STARTUP_DIAGNOSTIC_LINE_PATTERN.test(line));
   const tailLines = (startupLines.length > 0 ? startupLines : lines).slice(-8);
   const tailText = tailLines.join("\n");
 
@@ -40,7 +43,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
   }
 
   const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
-  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_OR_OVERLOAD_PATTERN.test(content);
+  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_PATTERN.test(content);
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
   const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
 

From 2da23d0c78b57bd7bb53028d3f2c010b1cc1c761 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:59:26 +0000
Subject: [PATCH 05/14] chore: polish diagnostics constants and readability

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       | 3 ++-
 actions/setup/js/log_parser_bootstrap.cjs | 8 ++++++--
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index e9b630b278d..8c3fcc2eeaf 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -78,6 +78,7 @@ const MAX_TURNS_EXIT_PATTERN = /"subtype"\s*:\s*"error_max_turns"/;
 // this path must not be retried via --continue (fall back to a fresh run if budget remains).
 const NO_DEFERRED_MARKER_PATTERN = /No deferred tool marker found/i;
 const SIGNAL_TERMINATION_EXIT_CODES = new Set([137, 143]);
+const MAX_STARTUP_RETRIES = 2;
 
 /**
  * Emit a timestamped diagnostic log line to stderr.
@@ -121,7 +122,7 @@ function resolveStartupRetryLimit(env = process.env, logger = log) {
     logger(`invalid GH_AW_CLAUDE_STARTUP_RETRIES='${raw}' (not finite); using default startup retries=1`);
     return 1;
   }
-  const clamped = Math.min(Math.max(parsed, 0), 2);
+  const clamped = Math.min(Math.max(parsed, 0), MAX_STARTUP_RETRIES);
   if (clamped !== parsed) {
     logger(`GH_AW_CLAUDE_STARTUP_RETRIES=${parsed} out of range; clamped to ${clamped}`);
   }
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index b9270f746c7..874bccee69b 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -9,6 +9,7 @@ const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_e
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
 const CLAUDE_HTTP_5XX_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i;
+const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
 /**
  * Build startup diagnostics for Claude failures with no structured entries.
@@ -28,7 +29,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
     .map(line => line.trim())
     .filter(Boolean);
   const startupLines = lines.filter(line => line.includes("[claude-harness]") || STARTUP_DIAGNOSTIC_LINE_PATTERN.test(line));
-  const tailLines = (startupLines.length > 0 ? startupLines : lines).slice(-8);
+  const tailLines = (startupLines.length > 0 ? startupLines : lines).slice(-MAX_DIAGNOSTIC_TAIL_LINES);
   const tailText = tailLines.join("\n");
 
   let exitCode = "unknown";
@@ -43,7 +44,10 @@ function buildClaudeStartupDiagnostics(rawContent) {
   }
 
   const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
-  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_PATTERN.test(content);
+  const matchedRateLimit = CLAUDE_RATE_LIMIT_PATTERN.test(content);
+  const matchedOverload = CLAUDE_OVERLOAD_PATTERN.test(content);
+  const matchedHTTP5xx = CLAUDE_HTTP_5XX_PATTERN.test(content);
+  const aiCreditsRateLimitError = matchedRateLimit || matchedOverload || matchedHTTP5xx;
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
   const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
 

From 237a3941af21073af6984e209f1c0ade81cff5e9 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 16:03:22 +0000
Subject: [PATCH 06/14] chore: tighten startup parsing and logging efficiency

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       |  4 +++-
 actions/setup/js/log_parser_bootstrap.cjs | 27 ++++++++++++++---------
 2 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 8c3fcc2eeaf..17dd9a36640 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -554,7 +554,9 @@ async function main() {
       if (startupRetriesUsed < startupRetryLimit) {
         startupRetriesUsed++;
         useContinueOnRetry = false;
-        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${attempt + 2} of ${maxRetries + 1} total attempts)`);
+        const nextAttempt = attempt + 2;
+        const totalAttempts = maxRetries + 1;
+        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttempt} of ${totalAttempts} total attempts)`);
         continue;
       }
       log(
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 874bccee69b..3d6bfa7d2c9 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -33,21 +33,28 @@ function buildClaudeStartupDiagnostics(rawContent) {
   const tailText = tailLines.join("\n");
 
   let exitCode = "unknown";
-  const doneMatches = [...content.matchAll(/done:\s*exitCode=(\d+)/g)];
-  if (doneMatches.length > 0) {
-    exitCode = doneMatches[doneMatches.length - 1][1];
+  const donePrefix = "done: exitCode=";
+  const doneIdx = content.lastIndexOf(donePrefix);
+  if (doneIdx !== -1) {
+    const doneTail = content.slice(doneIdx + donePrefix.length);
+    const doneCode = doneTail.match(/^(\d+)/);
+    if (doneCode) {
+      exitCode = doneCode[1];
+    }
   } else {
-    const attemptMatches = [...content.matchAll(/attempt\s+\d+\s+failed:\s+exitCode=(\d+)/g)];
-    if (attemptMatches.length > 0) {
-      exitCode = attemptMatches[attemptMatches.length - 1][1];
+    const failedPrefix = "failed: exitCode=";
+    const failedIdx = content.lastIndexOf(failedPrefix);
+    if (failedIdx !== -1) {
+      const failedTail = content.slice(failedIdx + failedPrefix.length);
+      const failedCode = failedTail.match(/^(\d+)/);
+      if (failedCode) {
+        exitCode = failedCode[1];
+      }
     }
   }
 
   const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
-  const matchedRateLimit = CLAUDE_RATE_LIMIT_PATTERN.test(content);
-  const matchedOverload = CLAUDE_OVERLOAD_PATTERN.test(content);
-  const matchedHTTP5xx = CLAUDE_HTTP_5XX_PATTERN.test(content);
-  const aiCreditsRateLimitError = matchedRateLimit || matchedOverload || matchedHTTP5xx;
+  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_PATTERN.test(content);
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
   const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
 

From d8e40a894dbb15781558604ebb3b20348e508a1a Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 16:07:16 +0000
Subject: [PATCH 07/14] chore: apply review-driven readability refinements

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       | 6 ++----
 actions/setup/js/log_parser_bootstrap.cjs | 7 ++++---
 2 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 17dd9a36640..9f14baf641e 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -556,13 +556,11 @@ async function main() {
         useContinueOnRetry = false;
         const nextAttempt = attempt + 2;
         const totalAttempts = maxRetries + 1;
-        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttempt} of ${totalAttempts} total attempts)`);
+        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run (startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttempt} of ${totalAttempts} total attempts)`);
         continue;
       }
       log(
-        `attempt ${attempt + 1}: no output produced — not retrying` +
-          ` (startup retry budget exhausted: ${startupRetriesUsed}/${startupRetryLimit};` +
-          ` possible causes: binary not found, permission denied, auth failure, or silent startup crash)`
+        `attempt ${attempt + 1}: no output produced — not retrying (startup retry budget exhausted: ${startupRetriesUsed}/${startupRetryLimit}; possible causes: binary not found, permission denied, auth failure, or silent startup crash)`
       );
     }
 
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 3d6bfa7d2c9..ad542020b54 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,7 +7,7 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-const CLAUDE_HTTP_5XX_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
@@ -29,7 +29,8 @@ function buildClaudeStartupDiagnostics(rawContent) {
     .map(line => line.trim())
     .filter(Boolean);
   const startupLines = lines.filter(line => line.includes("[claude-harness]") || STARTUP_DIAGNOSTIC_LINE_PATTERN.test(line));
-  const tailLines = (startupLines.length > 0 ? startupLines : lines).slice(-MAX_DIAGNOSTIC_TAIL_LINES);
+  const diagnosticLines = startupLines.length > 0 ? startupLines : lines;
+  const tailLines = diagnosticLines.slice(-MAX_DIAGNOSTIC_TAIL_LINES);
   const tailText = tailLines.join("\n");
 
   let exitCode = "unknown";
@@ -54,7 +55,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
   }
 
   const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
-  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_PATTERN.test(content);
+  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_STATUS_PATTERN.test(content);
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
   const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
 

From 9cd8ddd03a7463f0abea1c8d2e09cd5467008678 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 16:11:19 +0000
Subject: [PATCH 08/14] chore: clarify transient classification semantics

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs       |  1 +
 actions/setup/js/log_parser_bootstrap.cjs | 11 +++++++----
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 9f14baf641e..f79c2201121 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -554,6 +554,7 @@ async function main() {
       if (startupRetriesUsed < startupRetryLimit) {
         startupRetriesUsed++;
         useContinueOnRetry = false;
+        // attempt is 0-based; logs are 1-based, so "next attempt" is +2.
         const nextAttempt = attempt + 2;
         const totalAttempts = maxRetries + 1;
         log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run (startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttempt} of ${totalAttempts} total attempts)`);
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index ad542020b54..a005ab56dee 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,7 +7,7 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d\d\b/i;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d{2}\b/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
@@ -18,6 +18,7 @@ const MAX_DIAGNOSTIC_TAIL_LINES = 8;
  *   exitCode: string,
  *   inferenceAccessError: boolean,
  *   aiCreditsRateLimitError: boolean,
+ *   transientInferenceAvailabilityError: boolean,
  *   summaryLine: string,
  *   summaryMarkdown: string
  * }}
@@ -55,7 +56,8 @@ function buildClaudeStartupDiagnostics(rawContent) {
   }
 
   const inferenceAccessError = INFERENCE_ACCESS_ERROR_PATTERN.test(content);
-  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content) || CLAUDE_HTTP_5XX_STATUS_PATTERN.test(content);
+  const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content);
+  const transientInferenceAvailabilityError = aiCreditsRateLimitError || CLAUDE_HTTP_5XX_STATUS_PATTERN.test(content);
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
   const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
 
@@ -63,6 +65,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
     exitCode,
     inferenceAccessError,
     aiCreditsRateLimitError,
+    transientInferenceAvailabilityError,
     summaryLine,
     summaryMarkdown,
   };
@@ -419,8 +422,8 @@ async function runLogParser(options) {
           core.setOutput("ai_credits_rate_limit_error", "true");
         }
 
-        const errorCode = diagnostics.inferenceAccessError || diagnostics.aiCreditsRateLimitError ? ERR_API : ERR_CONFIG;
-        const failureKind = diagnostics.inferenceAccessError || diagnostics.aiCreditsRateLimitError ? "transient inference availability signal detected" : "startup/configuration failure detected";
+        const errorCode = diagnostics.inferenceAccessError || diagnostics.transientInferenceAvailabilityError ? ERR_API : ERR_CONFIG;
+        const failureKind = diagnostics.inferenceAccessError || diagnostics.transientInferenceAvailabilityError ? "transient inference availability signal detected" : "startup/configuration failure detected";
 
         core.setFailed(`${errorCode}: Claude execution failed: no structured log entries were produced. ${diagnostics.summaryLine} ${failureKind}.`);
       }

From ee6d3be92010abc44c2140f67e54e08d1ddfccee Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 8 Jul 2026 16:15:21 +0000
Subject: [PATCH 09/14] chore: refine regex precision and naming clarity

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs            | 4 ++--
 actions/setup/js/log_parser_bootstrap.cjs      | 4 ++--
 actions/setup/js/log_parser_bootstrap.test.cjs | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index f79c2201121..12841af0af5 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -555,9 +555,9 @@ async function main() {
         startupRetriesUsed++;
         useContinueOnRetry = false;
         // attempt is 0-based; logs are 1-based, so "next attempt" is +2.
-        const nextAttempt = attempt + 2;
+        const nextAttemptNumber = attempt + 2;
         const totalAttempts = maxRetries + 1;
-        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run (startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttempt} of ${totalAttempts} total attempts)`);
+        log(`attempt ${attempt + 1}: no output produced — retrying startup as fresh run ` + `(startup retry ${startupRetriesUsed}/${startupRetryLimit}, next attempt ${nextAttemptNumber} of ${totalAttempts} total attempts)`);
         continue;
       }
       log(
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index a005ab56dee..f9e3998612a 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,8 +7,8 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP|status|error)[^\n]{0,60}\b5\d{2}\b/i;
-const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|5\d\d|overloaded|inference)/i;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|error(?:\s+code)?\s*[:=]?\s*5\d{2}\b)/i;
+const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
 /**
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index 8eff4688e3c..b41ea1a3aaf 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -291,7 +291,7 @@ describe("log_parser_bootstrap.cjs", () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");
           try {
-            fs.writeFileSync(logFile, `[claude-harness] attempt 1 failed: exitCode=1 isRateLimitError=true hasOutput=false\n[claude-harness] done: exitCode=1\nAPI Error: Request rejected (429)`);
+            fs.writeFileSync(logFile, `[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] done: exitCode=1\nAPI Error: Request rejected (429)`);
             process.env.GH_AW_AGENT_OUTPUT = logFile;
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });

From edf96e7dbfa095e1d3f98de8bd8d12dc273dea71 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:14:41 +0000
Subject: [PATCH 10/14] fix: harden Claude startup diagnostics summaries

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.cjs           |  1 +
 actions/setup/js/claude_harness.test.cjs      | 20 +++++-
 actions/setup/js/log_parser_bootstrap.cjs     | 22 ++++--
 .../setup/js/log_parser_bootstrap.test.cjs    | 71 +++++++++++++++++++
 4 files changed, 107 insertions(+), 7 deletions(-)

diff --git a/actions/setup/js/claude_harness.cjs b/actions/setup/js/claude_harness.cjs
index 12841af0af5..3ddf5f2daa7 100644
--- a/actions/setup/js/claude_harness.cjs
+++ b/actions/setup/js/claude_harness.cjs
@@ -554,6 +554,7 @@ async function main() {
       if (startupRetriesUsed < startupRetryLimit) {
         startupRetriesUsed++;
         useContinueOnRetry = false;
+        delay = initialDelayMs;
         // attempt is 0-based; logs are 1-based, so "next attempt" is +2.
         const nextAttemptNumber = attempt + 2;
         const totalAttempts = maxRetries + 1;
diff --git a/actions/setup/js/claude_harness.test.cjs b/actions/setup/js/claude_harness.test.cjs
index c9d09dbf35d..290bd136e22 100644
--- a/actions/setup/js/claude_harness.test.cjs
+++ b/actions/setup/js/claude_harness.test.cjs
@@ -31,7 +31,7 @@ function makeHarnessTempDir(name) {
   return fs.mkdtempSync(path.join(agentTempDir, name));
 }
 
-function runHarnessWithStub({ stubScript, prompt = "fix the bug", extraArgs = [] }) {
+function runHarnessWithStub({ stubScript, prompt = "fix the bug", extraArgs = [], extraEnv = {} }) {
   const tempDir = makeHarnessTempDir("claude-harness-");
   const stubPath = path.join(tempDir, "stub.cjs");
   const promptPath = path.join(tempDir, "prompt.txt");
@@ -41,7 +41,7 @@ function runHarnessWithStub({ stubScript, prompt = "fix the bug", extraArgs = []
 
   const result = spawnSync(process.execPath, ["claude_harness.cjs", process.execPath, stubPath, "--print", ...extraArgs, "--prompt-file", promptPath], {
     cwd: path.dirname(require.resolve("./claude_harness.cjs")),
-    env: { ...process.env, CLAUDE_HARNESS_STUB_CALLS: callsPath },
+    env: { ...process.env, ...extraEnv, CLAUDE_HARNESS_STUB_CALLS: callsPath },
     encoding: "utf8",
     timeout: 45000,
   });
@@ -467,6 +467,22 @@ process.exit(0);
       expect(result.stderr).toContain("no output produced — retrying startup as fresh run");
     });
 
+    it("does not retry no-output startup failure when GH_AW_CLAUDE_STARTUP_RETRIES=0", () => {
+      const stubScript = `
+const fs = require("fs");
+const callsPath = process.env.CLAUDE_HARNESS_STUB_CALLS;
+fs.appendFileSync(callsPath, JSON.stringify({ args: process.argv.slice(2) }) + "\\n", "utf8");
+process.exit(1);
+`;
+      const { result, calls } = runHarnessWithStub({
+        stubScript,
+        extraEnv: { GH_AW_CLAUDE_STARTUP_RETRIES: "0" },
+      });
+      expect(result.status).not.toBe(0);
+      expect(calls.length).toBe(1);
+      expect(result.stderr).toContain("startup retry budget exhausted: 0/0");
+    });
+
     it("returns true for normal partial-execution retry", () => {
       const result = shouldRetryWithContinue({
         attempt: 0,
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index f9e3998612a..b4f9fd0e153 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,7 +7,7 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|error(?:\s+code)?\s*[:=]?\s*5\d{2}\b)/i;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|(?:http|fetch|request)\s+(?:failed|error)[^\n]*\b5\d{2}\b)/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
@@ -30,9 +30,9 @@ function buildClaudeStartupDiagnostics(rawContent) {
     .map(line => line.trim())
     .filter(Boolean);
   const startupLines = lines.filter(line => line.includes("[claude-harness]") || STARTUP_DIAGNOSTIC_LINE_PATTERN.test(line));
-  const diagnosticLines = startupLines.length > 0 ? startupLines : lines;
-  const tailLines = diagnosticLines.slice(-MAX_DIAGNOSTIC_TAIL_LINES);
+  const tailLines = startupLines.slice(-MAX_DIAGNOSTIC_TAIL_LINES);
   const tailText = tailLines.join("\n");
+  const safeTailText = escapeHtml(tailText);
 
   let exitCode = "unknown";
   const donePrefix = "done: exitCode=";
@@ -59,7 +59,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
   const aiCreditsRateLimitError = CLAUDE_RATE_LIMIT_PATTERN.test(content) || CLAUDE_OVERLOAD_PATTERN.test(content);
   const transientInferenceAvailabilityError = aiCreditsRateLimitError || CLAUDE_HTTP_5XX_STATUS_PATTERN.test(content);
   const summaryLine = `Claude startup failed before structured logging (exitCode=${exitCode}).`;
-  const summaryMarkdown = tailText ? `<details><summary>Claude startup diagnostics</summary>\n\n\`\`\`text\n${tailText}\n\`\`\`\n</details>` : "";
+  const summaryMarkdown = safeTailText ? `<details><summary>Claude startup diagnostics</summary>\n\n<pre><code>${safeTailText}</code></pre>\n</details>` : "";
 
   return {
     exitCode,
@@ -71,6 +71,14 @@ function buildClaudeStartupDiagnostics(rawContent) {
   };
 }
 
+/**
+ * @param {string} text
+ * @returns {string}
+ */
+function escapeHtml(text) {
+  return text.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;");
+}
+
 /**
  * Bootstrap helper for log parser entry points.
  * Handles common logic for environment variable lookup, file existence checks,
@@ -423,7 +431,11 @@ async function runLogParser(options) {
         }
 
         const errorCode = diagnostics.inferenceAccessError || diagnostics.transientInferenceAvailabilityError ? ERR_API : ERR_CONFIG;
-        const failureKind = diagnostics.inferenceAccessError || diagnostics.transientInferenceAvailabilityError ? "transient inference availability signal detected" : "startup/configuration failure detected";
+        const failureKind = diagnostics.inferenceAccessError
+          ? "inference access denied by policy"
+          : diagnostics.transientInferenceAvailabilityError
+            ? "transient inference availability signal detected"
+            : "startup/configuration failure detected";
 
         core.setFailed(`${errorCode}: Claude execution failed: no structured log entries were produced. ${diagnostics.summaryLine} ${failureKind}.`);
       }
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index b41ea1a3aaf..1332b415d89 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -247,6 +247,41 @@ describe("log_parser_bootstrap.cjs", () => {
             fs.rmdirSync(tmpDir);
           }
         }),
+        it("should omit Claude startup diagnostics tail when no startup lines are matched", async () => {
+          const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
+          const logFile = path.join(tmpDir, "test.log");
+          try {
+            fs.writeFileSync(logFile, "secret-token-value\nplain stdout line");
+            process.env.GH_AW_AGENT_OUTPUT = logFile;
+            delete process.env.GH_AW_SAFE_OUTPUTS;
+            const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
+            await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
+            expect(mockCore.summary.addRaw).toHaveBeenCalledTimes(1);
+            expect(mockCore.summary.addRaw.mock.calls.flat().join("\n")).not.toContain("secret-token-value");
+          } finally {
+            fs.unlinkSync(logFile);
+            delete process.env.GH_AW_AGENT_OUTPUT;
+            fs.rmdirSync(tmpDir);
+          }
+        }),
+        it("should escape Claude startup diagnostics summary content", async () => {
+          const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
+          const logFile = path.join(tmpDir, "test.log");
+          try {
+            fs.writeFileSync(logFile, "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1");
+            process.env.GH_AW_AGENT_OUTPUT = logFile;
+            delete process.env.GH_AW_SAFE_OUTPUTS;
+            const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
+            await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
+            const diagnosticsSummary = mockCore.summary.addRaw.mock.calls[1][0];
+            expect(diagnosticsSummary).toContain("<pre><code>");
+            expect(diagnosticsSummary).toContain("&lt;pre&gt;&amp;```&lt;/pre&gt;");
+          } finally {
+            fs.unlinkSync(logFile);
+            delete process.env.GH_AW_AGENT_OUTPUT;
+            fs.rmdirSync(tmpDir);
+          }
+        }),
         it("should not print 'parsed successfully' for Claude when logEntries is empty", async () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");
@@ -306,6 +341,42 @@ describe("log_parser_bootstrap.cjs", () => {
             fs.rmdirSync(tmpDir);
           }
         }),
+        it("should classify Claude empty-log startup inference-access denials separately", async () => {
+          const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
+          const logFile = path.join(tmpDir, "test.log");
+          try {
+            fs.writeFileSync(logFile, `[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\nAccess denied by policy settings\n[claude-harness] done: exitCode=1`);
+            process.env.GH_AW_AGENT_OUTPUT = logFile;
+            delete process.env.GH_AW_SAFE_OUTPUTS;
+            const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
+            await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
+            expect(mockCore.setFailed).toHaveBeenCalledWith(`${ERR_API}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=1). inference access denied by policy.`);
+            expect(mockCore.setOutput).toHaveBeenCalledWith("inference_access_error", "true");
+            expect(mockCore.setOutput).not.toHaveBeenCalledWith("ai_credits_rate_limit_error", "true");
+          } finally {
+            fs.unlinkSync(logFile);
+            delete process.env.GH_AW_AGENT_OUTPUT;
+            fs.rmdirSync(tmpDir);
+          }
+        }),
+        it("should not classify non-HTTP 500 text as transient inference availability", async () => {
+          const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
+          const logFile = path.join(tmpDir, "test.log");
+          try {
+            fs.writeFileSync(logFile, `[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\nerror code 500\n[claude-harness] done: exitCode=1`);
+            process.env.GH_AW_AGENT_OUTPUT = logFile;
+            delete process.env.GH_AW_SAFE_OUTPUTS;
+            const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });
+            await runLogParser({ parseLog: mockParseLog, parserName: "Claude" });
+            expect(mockCore.setFailed).toHaveBeenCalledWith(
+              `${ERR_CONFIG}: Claude execution failed: no structured log entries were produced. Claude startup failed before structured logging (exitCode=1). startup/configuration failure detected.`
+            );
+          } finally {
+            fs.unlinkSync(logFile);
+            delete process.env.GH_AW_AGENT_OUTPUT;
+            fs.rmdirSync(tmpDir);
+          }
+        }),
         it("should treat logEntries: null as missing entries for Claude guardrail (safe outputs → warning)", async () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");

From 12828fbfbca3bbfe904d53ca932be03f93305cc6 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:18:13 +0000
Subject: [PATCH 11/14] test: cover Claude startup diagnostics edge cases

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/log_parser_bootstrap.cjs      | 2 +-
 actions/setup/js/log_parser_bootstrap.test.cjs | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index b4f9fd0e153..c328ca687fb 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,7 +7,7 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|(?:http|fetch|request)\s+(?:failed|error)[^\n]*\b5\d{2}\b)/i;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|(?:http|fetch|request)\s+(?:failed|error)[^\n]*?\b5\d{2}\b)/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index 1332b415d89..b860c6fdfa2 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -267,8 +267,9 @@ describe("log_parser_bootstrap.cjs", () => {
         it("should escape Claude startup diagnostics summary content", async () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");
+          const escapedTailLog = "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1";
           try {
-            fs.writeFileSync(logFile, "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1");
+            fs.writeFileSync(logFile, escapedTailLog);
             process.env.GH_AW_AGENT_OUTPUT = logFile;
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });

From a120adcadf3de7454fc47e154cc148e71505cd8f Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:21:43 +0000
Subject: [PATCH 12/14] chore: tighten Claude diagnostics matching and escaping

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/log_parser_bootstrap.cjs      | 6 +++++-
 actions/setup/js/log_parser_bootstrap.test.cjs | 4 ++--
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index c328ca687fb..2b94b184d09 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,6 +7,10 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
+// Treat only clearly HTTP-scoped 5xx strings as transient inference availability:
+//  - protocol/status lines such as "HTTP/1.1 503"
+//  - structured status-code fields such as "status code: 502"
+//  - request/fetch/http error prose that includes a 5xx code
 const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|(?:http|fetch|request)\s+(?:failed|error)[^\n]*?\b5\d{2}\b)/i;
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
@@ -76,7 +80,7 @@ function buildClaudeStartupDiagnostics(rawContent) {
  * @returns {string}
  */
 function escapeHtml(text) {
-  return text.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;");
+  return text.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll('"', "&quot;").replaceAll("'", "&#39;");
 }
 
 /**
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index b860c6fdfa2..916a5f38b48 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -267,9 +267,9 @@ describe("log_parser_bootstrap.cjs", () => {
         it("should escape Claude startup diagnostics summary content", async () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");
-          const escapedTailLog = "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1";
+          const rawTailLogWithHtml = "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1";
           try {
-            fs.writeFileSync(logFile, escapedTailLog);
+            fs.writeFileSync(logFile, rawTailLogWithHtml);
             process.env.GH_AW_AGENT_OUTPUT = logFile;
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });

From dafe82eb4bb33ba508f06075d20ce9eba3c78c2f Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:25:29 +0000
Subject: [PATCH 13/14] refactor: clarify Claude startup diagnostics matching

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/claude_harness.test.cjs       |  3 ++-
 actions/setup/js/log_parser_bootstrap.cjs      | 12 +++++++-----
 actions/setup/js/log_parser_bootstrap.test.cjs |  4 ++--
 3 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/actions/setup/js/claude_harness.test.cjs b/actions/setup/js/claude_harness.test.cjs
index 290bd136e22..c0767aee072 100644
--- a/actions/setup/js/claude_harness.test.cjs
+++ b/actions/setup/js/claude_harness.test.cjs
@@ -478,8 +478,9 @@ process.exit(1);
         stubScript,
         extraEnv: { GH_AW_CLAUDE_STARTUP_RETRIES: "0" },
       });
-      expect(result.status).not.toBe(0);
+      expect(result.status).toBe(1);
       expect(calls.length).toBe(1);
+      expect(calls[0].args).toContain("fix the bug");
       expect(result.stderr).toContain("startup retry budget exhausted: 0/0");
     });
 
diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 2b94b184d09..133604f52df 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,11 +7,10 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
-// Treat only clearly HTTP-scoped 5xx strings as transient inference availability:
-//  - protocol/status lines such as "HTTP/1.1 503"
-//  - structured status-code fields such as "status code: 502"
-//  - request/fetch/http error prose that includes a 5xx code
-const CLAUDE_HTTP_5XX_STATUS_PATTERN = /(?:HTTP(?:\/\d\.\d)?\s+5\d{2}\b|status(?:\s+code)?\s*[:=]\s*5\d{2}\b|(?:http|fetch|request)\s+(?:failed|error)[^\n]*?\b5\d{2}\b)/i;
+const CLAUDE_HTTP_5XX_PROTOCOL_PATTERN = /HTTP(?:\/\d\.\d)?\s+5\d{2}\b/;
+const CLAUDE_HTTP_5XX_STATUS_FIELD_PATTERN = /status(?:\s+code)?\s*[:=]\s*5\d{2}\b/;
+const CLAUDE_HTTP_5XX_REQUEST_ERROR_PATTERN = /(?:http|fetch|request)\s+(?:failed|error)[^\n]*?\b5\d{2}\b/;
+const CLAUDE_HTTP_5XX_STATUS_PATTERN = new RegExp([CLAUDE_HTTP_5XX_PROTOCOL_PATTERN.source, CLAUDE_HTTP_5XX_STATUS_FIELD_PATTERN.source, CLAUDE_HTTP_5XX_REQUEST_ERROR_PATTERN.source].join("|"), "i");
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
 const MAX_DIAGNOSTIC_TAIL_LINES = 8;
 
@@ -34,6 +33,9 @@ function buildClaudeStartupDiagnostics(rawContent) {
     .map(line => line.trim())
     .filter(Boolean);
   const startupLines = lines.filter(line => line.includes("[claude-harness]") || STARTUP_DIAGNOSTIC_LINE_PATTERN.test(line));
+  // Intentionally avoid falling back to arbitrary raw log lines here. For empty-structured-log
+  // failures, the unmatched tail can contain unrelated stdout/stderr, prompts, or secrets; when
+  // no startup/diagnostic lines are recognized, we emit no tail section rather than risk leaking it.
   const tailLines = startupLines.slice(-MAX_DIAGNOSTIC_TAIL_LINES);
   const tailText = tailLines.join("\n");
   const safeTailText = escapeHtml(tailText);
diff --git a/actions/setup/js/log_parser_bootstrap.test.cjs b/actions/setup/js/log_parser_bootstrap.test.cjs
index 916a5f38b48..bd1d5ffb77b 100644
--- a/actions/setup/js/log_parser_bootstrap.test.cjs
+++ b/actions/setup/js/log_parser_bootstrap.test.cjs
@@ -267,9 +267,9 @@ describe("log_parser_bootstrap.cjs", () => {
         it("should escape Claude startup diagnostics summary content", async () => {
           const tmpDir = fs.mkdtempSync(path.join(__dirname, "test-"));
           const logFile = path.join(tmpDir, "test.log");
-          const rawTailLogWithHtml = "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1";
+          const logContentWithHtmlChars = "[claude-harness] attempt 1 failed: exitCode=1 hasOutput=false\n[claude-harness] stderr: <pre>&```</pre>\n[claude-harness] done: exitCode=1";
           try {
-            fs.writeFileSync(logFile, rawTailLogWithHtml);
+            fs.writeFileSync(logFile, logContentWithHtmlChars);
             process.env.GH_AW_AGENT_OUTPUT = logFile;
             delete process.env.GH_AW_SAFE_OUTPUTS;
             const mockParseLog = vi.fn().mockReturnValue({ markdown: "## Result\n", mcpFailures: [], maxTurnsHit: false, logEntries: [] });

From 833c290ae74c6549d69510d7c0dd1ba4fae6dbc7 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:29:00 +0000
Subject: [PATCH 14/14] chore: document Claude startup 5xx matchers

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
---
 actions/setup/js/log_parser_bootstrap.cjs | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/actions/setup/js/log_parser_bootstrap.cjs b/actions/setup/js/log_parser_bootstrap.cjs
index 133604f52df..32c60de6965 100644
--- a/actions/setup/js/log_parser_bootstrap.cjs
+++ b/actions/setup/js/log_parser_bootstrap.cjs
@@ -7,8 +7,11 @@ const { ERR_API, ERR_CONFIG, ERR_VALIDATION } = require("./error_codes.cjs");
 const INFERENCE_ACCESS_ERROR_PATTERN = /Access denied by policy settings|invalid access to inference/i;
 const CLAUDE_RATE_LIMIT_PATTERN = /rate_limit_error|429 Too Many Requests|"api_error_status"\s*:\s*429|request rejected \(429\)|rate limit/i;
 const CLAUDE_OVERLOAD_PATTERN = /overloaded_error|"overloaded"/i;
+/** Matches HTTP protocol/status lines such as "HTTP/1.1 503". */
 const CLAUDE_HTTP_5XX_PROTOCOL_PATTERN = /HTTP(?:\/\d\.\d)?\s+5\d{2}\b/;
+/** Matches structured status fields such as "status: 502" or "status code=500". */
 const CLAUDE_HTTP_5XX_STATUS_FIELD_PATTERN = /status(?:\s+code)?\s*[:=]\s*5\d{2}\b/;
+/** Matches request/fetch/http error prose that carries an explicit 5xx code. */
 const CLAUDE_HTTP_5XX_REQUEST_ERROR_PATTERN = /(?:http|fetch|request)\s+(?:failed|error)[^\n]*?\b5\d{2}\b/;
 const CLAUDE_HTTP_5XX_STATUS_PATTERN = new RegExp([CLAUDE_HTTP_5XX_PROTOCOL_PATTERN.source, CLAUDE_HTTP_5XX_STATUS_FIELD_PATTERN.source, CLAUDE_HTTP_5XX_REQUEST_ERROR_PATTERN.source].join("|"), "i");
 const STARTUP_DIAGNOSTIC_LINE_PATTERN = /(?:ERR_|Error:|CAPIError|Authentication failed|rate[_ -]?limit|429|\b5\d{2}\b|overloaded|inference)/i;
@@ -82,7 +85,8 @@ function buildClaudeStartupDiagnostics(rawContent) {
  * @returns {string}
  */
 function escapeHtml(text) {
-  return text.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll('"', "&quot;").replaceAll("'", "&#39;");
+  const htmlEntities = { '"': "&quot;", "'": "&#39;", "&": "&amp;", "<": "&lt;", ">": "&gt;" };
+  return text.replace(/["'&<>]/g, char => htmlEntities[char]);
 }
 
 /**