ci failures refactor

Author: Marco Damonte

action.yml

@@ -175,18 +175,21 @@ runs:
         EJ_AUTH_GITHUB_TOKEN: ${{ steps.prepare.outputs.EJ_AUTH_GITHUB_TOKEN  }}
         GH_TOKEN: ${{ steps.prepare.outputs.EJ_AUTH_GITHUB_TOKEN }}
         MULTI_RUNS: true
-        MATTERHORN_DEBUG_LOG: false
+        MATTERHORN_DEBUG_LOG: true
         WORKING_DIR: ${{ inputs.junie_work_dir }}
         MODEL: ${{ inputs.model }}
       run: |
         MODEL_FLAG=""
         if [ -n "${MODEL}" ]; then
           MODEL_FLAG="--model=${MODEL}"
         fi
-        OUTPUT=$(junie --cache-dir="${WORKING_DIR}" --input-format="json" --output-format="json" --auth="${CLI_TOKEN}" ${MODEL_FLAG} <<<"${JUNIE_JSON_TASK}")
-        echo "json_output<<EOF" >> $GITHUB_OUTPUT
-        echo "$OUTPUT" >> $GITHUB_OUTPUT
-        echo "EOF" >> $GITHUB_OUTPUT
+        # Ensure working directory exists
+        mkdir -p "${WORKING_DIR}"
+        # Set output file path FIRST (so it's available even if Junie fails)
+        OUTPUT_FILE="${WORKING_DIR}/junie-output.json"
+        echo "json_output_file=${OUTPUT_FILE}" >> $GITHUB_OUTPUT
+        # Write output to file to avoid ARG_MAX limit with large outputs
+        junie --cache-dir="${WORKING_DIR}" --input-format="json" --output-format="json" --auth="${CLI_TOKEN}" ${MODEL_FLAG} <<<"${JUNIE_JSON_TASK}" > "${OUTPUT_FILE}"
 
     - name: Handle Junie results
       if: always() && steps.prepare.outputs.SHOULD_SKIP != 'true'
@@ -195,7 +198,7 @@ runs:
       run: |
         bun run ${GITHUB_ACTION_PATH}/src/entrypoints/handle-results.ts
       env:
-        JSON_JUNIE_OUTPUT: ${{ steps.junie-run.outputs.json_output }}
+        JSON_JUNIE_OUTPUT_FILE: ${{ steps.junie-run.outputs.json_output_file }}
         PARSED_CONTEXT: ${{ steps.prepare.outputs.PARSED_CONTEXT }}
         INIT_COMMENT_ID: ${{ steps.prepare.outputs.INIT_COMMENT_ID }}
         WORKING_BRANCH: ${{ steps.prepare.outputs.WORKING_BRANCH }}
@@ -282,7 +285,7 @@ runs:
         COMMIT_SHA: ${{ steps.commit.outputs.commit_long_sha }}
         JUNIE_TITLE: ${{ steps.junie-run-results.outputs.JUNIE_TITLE }}
         JUNIE_SUMMARY: ${{ steps.junie-run-results.outputs.JUNIE_SUMMARY }}
-        JSON_JUNIE_OUTPUT: ${{ steps.junie-run.outputs.json_output }}
+        JSON_JUNIE_OUTPUT_FILE: ${{ steps.junie-run.outputs.json_output_file }}
         JIRA_BASE_URL: ${{ inputs.jira_base_url }}
         JIRA_EMAIL: ${{ inputs.jira_email }}
         JIRA_API_TOKEN: ${{ inputs.jira_api_token }}
@@ -306,7 +309,7 @@ runs:
         COMMIT_SHA: ${{ steps.commit.outputs.commit_long_sha }}
         JUNIE_TITLE: ${{ steps.junie-run-results.outputs.JUNIE_TITLE }}
         JUNIE_SUMMARY: ${{ steps.junie-run-results.outputs.JUNIE_SUMMARY }}
-        JSON_JUNIE_OUTPUT: ${{ steps.junie-run.outputs.json_output }}
+        JSON_JUNIE_OUTPUT_FILE: ${{ steps.junie-run.outputs.json_output_file }}
         JIRA_BASE_URL: ${{ inputs.jira_base_url }}
         JIRA_EMAIL: ${{ inputs.jira_email }}
         JIRA_API_TOKEN: ${{ inputs.jira_api_token }}

full_test_output.txt

@@ -16,6 +16,7 @@ export const ENV_VARS = {
     OVERRIDE_GITHUB_TOKEN: "OVERRIDE_GITHUB_TOKEN",
     DEFAULT_WORKFLOW_TOKEN: "DEFAULT_WORKFLOW_TOKEN",
     JSON_JUNIE_OUTPUT: "JSON_JUNIE_OUTPUT",
+    JSON_JUNIE_OUTPUT_FILE: "JSON_JUNIE_OUTPUT_FILE",
     JUNIE_WORKING_DIR: "JUNIE_WORKING_DIR",
     APP_TOKEN: "APP_TOKEN",
     TRIGGER_PHRASE: "TRIGGER_PHRASE",

full_test_output_v2.txt

@@ -22,6 +22,10 @@ export const CODE_REVIEW_ACTION = "code-review";
 
 export const CODE_REVIEW_TRIGGER_PHRASE_REGEXP = new RegExp(CODE_REVIEW_ACTION, 'i')
 
+export const FIX_CI_ACTION = "fix-ci";
+
+export const FIX_CI_TRIGGER_PHRASE_REGEXP = new RegExp(FIX_CI_ACTION, 'i');
+
 export const JIRA_EVENT_ACTION = "jira_event";
 
 export const WORKING_BRANCH_PREFIX = "junie/";
@@ -79,6 +83,72 @@ Additional instructions:
 `;
 }
 
+export function createFixCIFailuresPrompt(diffPoint: string): string {
+    const diffCommand = `gh pr diff ${diffPoint}`
+    return `
+Your task is to analyze CI failures and suggest fixes WITHOUT implementing them. Follow these steps:
+
+1. Gather Information
+   - Use the 'get_pr_failed_checks_info' tool to retrieve information about failed CI/CD checks.
+   - Read the Pull Request diff by using \`${diffCommand} | grep "^diff --git"\`. Do not write the diff to file.
+
+2. If NO failed checks were found:
+   - Submit ONLY the following message:
+   ---
+   ## ✅ CI Status
+   
+   No failed checks found for this PR. All CI checks have passed or are still running.
+   ---
+
+3. If failed checks WERE found, analyze each failure:
+   - Open and explore relevant source files to understand the context
+   - Identify the failing step and error message
+   - Determine the root cause (test failure, build error, linting issue, timeout, flaky test, etc.)
+   - Correlate the error with changes in the PR diff
+   - Determine if the failure is related to the PR diff or a pre-existing issue
+
+4. Provide Diagnosis. After exploration, submit your analysis following this template FAITHFULLY and without adding additional notes:
+---
+## 🔴 CI Failure Analysis
+
+**Failed Check:** [check name]
+**Failed Step:** [step name if identifiable]
+**Error Type:** [test failure / build error / lint error / timeout / other]
+
+### Error Details
+\`\`\`
+[relevant error message/stack trace - keep concise]
+\`\`\`
+
+### Root Cause
+[1-3 sentences explaining why this failed]
+
+### Correlation with PR Changes
+[Explain which files/changes in this PR likely caused the failure, or state if it appears unrelated]
+
+## 🔧 Suggested Fix
+
+### What needs to change
+[Clear description of the fix approach]
+
+### Files to modify
+- \`path/to/file\`: [what needs to change and why]
+
+### Code changes
+\`\`\`[language]
+// Suggested code snippet or pseudocode
+\`\`\`
+---
+
+Additional Instructions:
+1. Open files or search the project as necessary to understand context before providing your diagnosis.
+2. Do NOT run tests, build, or make any modifications to the codebase.
+3. If multiple checks failed, analyze each one separately in your answer.
+4. If the failure appears to be a flaky test or infrastructure issue (not related to PR changes), clearly state this.
+5. Be specific about file paths and line numbers when suggesting fixes: \`File.ts:Line: Comment\`.
+6. Do not use the 'post_inline_review_comment' tool. Suggest changes only as shown in the template above`;
+}
+
 /**
  * Creates a hidden marker for identifying Junie comments from a specific workflow.
  * This HTML comment is invisible to users but allows finding Junie comments

src/constants/environment.ts

@@ -4,10 +4,25 @@ import {JunieExecutionContext} from "../github/context";
 import {ActionType} from "./handle-results";
 import {ENV_VARS, OUTPUT_VARS} from "../constants/environment";
 import {formatJunieSummary} from "./format-summary";
-import {appendFileSync} from "fs";
+import {appendFileSync, readFileSync, existsSync} from "fs";
 import {buildGitHubApiClient} from "../github/api/client";
 import {handleStepError} from "../utils/error-handler";
 
+/**
+ * Reads Junie output from file (preferred) or environment variable (fallback)
+ * File-based approach avoids ARG_MAX limit issues with large outputs
+ */
+function readJunieOutputSafe(): string | undefined {
+    // First try to read from file (new approach)
+    const outputFile = process.env[ENV_VARS.JSON_JUNIE_OUTPUT_FILE];
+    if (outputFile && existsSync(outputFile)) {
+        return readFileSync(outputFile, 'utf-8');
+    }
+    
+    // Fallback to environment variable (legacy approach)
+    return process.env[ENV_VARS.JSON_JUNIE_OUTPUT];
+}
+
 /**
  * Writes feedback comment to GitHub issue/PR if initCommentId is available
  */
@@ -62,7 +77,7 @@ async function generateJobSummary(isJobFailed: boolean): Promise<void> {
     }
 
     // Try to get duration_ms from JSON_JUNIE_OUTPUT if available
-    const jsonOutput = process.env[ENV_VARS.JSON_JUNIE_OUTPUT];
+    const jsonOutput = readJunieOutputSafe();
     if (jsonOutput) {
         try {
             const parsed = JSON.parse(jsonOutput);

src/constants/github.ts

@@ -6,6 +6,7 @@ import {ENV_VARS, OUTPUT_VARS} from "../constants/environment";
 import {handleStepError} from "../utils/error-handler";
 import {isReviewOrCommentHasResolveConflictsTrigger} from "../github/validation/trigger";
 import {sanitizeJunieOutput} from "../utils/sanitizer";
+import {readFileSync, existsSync} from 'fs';
 
 export enum ActionType {
     WRITE_COMMENT = 'WRITE_COMMENT',
@@ -15,18 +16,40 @@ export enum ActionType {
     NOTHING = 'NOTHING'
 }
 
+/**
+ * Reads Junie output from file (preferred) or environment variable (fallback)
+ * File-based approach avoids ARG_MAX limit issues with large outputs
+ */
+function readJunieOutput(): string {
+    // First try to read from file (new approach)
+    const outputFile = process.env[ENV_VARS.JSON_JUNIE_OUTPUT_FILE];
+    if (outputFile && existsSync(outputFile)) {
+        console.log(`Reading Junie output from file: ${outputFile}`);
+        const content = readFileSync(outputFile, 'utf-8');
+        // Debug: log first 1000 chars of content to help diagnose issues
+        console.log(`Junie output content (first 1000 chars): ${content.substring(0, 1000)}`);
+        return content;
+    }
+    
+    // Fallback to environment variable (legacy approach)
+    const envOutput = process.env[ENV_VARS.JSON_JUNIE_OUTPUT];
+    if (envOutput) {
+        console.log('Reading Junie output from environment variable (legacy)');
+        return envOutput;
+    }
+    
+    throw new Error(
+        `❌ Failed to retrieve Junie execution results. ` +
+        `This could be due to:\n` +
+        `• Junie execution did not complete successfully\n` +
+        `• Junie output was empty or invalid\n` +
+        `Please check the Junie execution logs for details.`
+    );
+}
+
 export async function handleResults() {
     try {
-        const stringJunieJsonOutput = process.env[ENV_VARS.JSON_JUNIE_OUTPUT]
-        if (!stringJunieJsonOutput) {
-            throw new Error(
-                `❌ Failed to retrieve Junie execution results. ` +
-                `This could be due to:\n` +
-                `• Junie execution did not complete successfully\n` +
-                `• Junie output was empty or invalid\n` +
-                `Please check the Junie execution logs for details.`
-            );
-        }
+        const stringJunieJsonOutput = readJunieOutput();
         const junieJsonOutput = JSON.parse(stringJunieJsonOutput) as any
         const context = JSON.parse(process.env[OUTPUT_VARS.PARSED_CONTEXT]!) as JunieExecutionContext
         const isResolveConflict = context.inputs.resolveConflicts || isReviewOrCommentHasResolveConflictsTrigger(context)

src/entrypoints/give-feedback.ts

@@ -458,6 +458,16 @@ export function isIssuesAssignedEvent(
     return isIssuesEvent(context) && context.eventAction === "assigned";
 }
 
+/**
+ * Checks if the context is a workflow_run event triggered by a CI failure
+ */
+export function isWorkflowRunFailureEvent(
+    context: JunieExecutionContext,
+): context is AutomationEventContext & { payload: WorkflowRunEvent } {
+    return context.eventName === "workflow_run" && 
+           (context.payload as WorkflowRunEvent).workflow_run?.conclusion === "failure";
+}
+
 /**
  * Checks if the context is triggered by user interaction (comments, PR/issue events)
  */

src/entrypoints/handle-results.ts

@@ -4,16 +4,18 @@ import {
     isPullRequestEvent,
     isPullRequestReviewCommentEvent,
     isPullRequestReviewEvent,
+    isWorkflowRunFailureEvent,
     JunieExecutionContext
 } from "../context";
 import * as core from "@actions/core";
 import {BranchInfo} from "../operations/branch";
 import {
     isReviewOrCommentHasCodeReviewTrigger,
+    isReviewOrCommentHasFixCITrigger,
     isReviewOrCommentHasResolveConflictsTrigger
 } from "../validation/trigger";
 import {OUTPUT_VARS} from "../../constants/environment";
-import {CODE_REVIEW_ACTION, createCodeReviewPrompt} from "../../constants/github";
+import {CODE_REVIEW_ACTION, createCodeReviewPrompt, createFixCIFailuresPrompt, FIX_CI_ACTION} from "../../constants/github";
 import {Octokits} from "../api/client";
 import {NewGitHubPromptFormatter} from "./new-prompt-formatter";
 import {downloadAttachmentsAndRewriteText} from "./attachment-downloader";
@@ -75,12 +77,29 @@ export async function prepareJunieTask(
         const isCodeReviewInComment = isReviewOrCommentHasCodeReviewTrigger(context);
         const isCodeReview = isCodeReviewInPrompt || isCodeReviewInComment;
 
+        // Check if prompt contains FIX_CI_ACTION phrase, comment/review has fix-ci trigger, or workflow_run CI failure
+        const isFixCIInPrompt = customPrompt?.includes(FIX_CI_ACTION);
+        const isFixCIInComment = isReviewOrCommentHasFixCITrigger(context);
+        const isFixCIFromWorkflowFailure = isWorkflowRunFailureEvent(context);
+        const isFixCI = isFixCIInPrompt || isFixCIInComment || isFixCIFromWorkflowFailure;
+    
+        // Debug: log fix-ci detection status
+        console.log(`Fix-CI detection: inPrompt=${isFixCIInPrompt}, inComment=${isFixCIInComment}, fromWorkflowFailure=${isFixCIFromWorkflowFailure}, isFixCI=${isFixCI}`);
+
         let promptText: string;
         let finalCustomPrompt = customPrompt;
         if (issue && isCodeReview) {
             const branchName = branchInfo.prBaseBranch || branchInfo.baseBranch;
             const diffPoint = context.isPR ? String(context.entityNumber) : branchName;
             finalCustomPrompt = createCodeReviewPrompt(diffPoint);
+            console.log(`Using CODE REVIEW prompt for diffPoint: ${diffPoint}`);
+        } else if (isFixCI) {
+            // Fix CI can trigger with or without an associated PR/issue (e.g., workflow_run failure on a branch)
+            const branchName = branchInfo.prBaseBranch || branchInfo.baseBranch;
+            const diffPoint = context.isPR && context.entityNumber ? String(context.entityNumber) : branchName;
+            finalCustomPrompt = createFixCIFailuresPrompt(diffPoint);
+            console.log(`Using FIX-CI prompt for diffPoint: ${diffPoint}`);
+            console.log(`Fix-CI prompt preview (first 200 chars): ${finalCustomPrompt.substring(0, 200)}`);
         }
         promptText = await formatter.generatePrompt(context, fetchedData, finalCustomPrompt, context.inputs.attachGithubContextToCustomPrompt);
 

src/github/context.ts

@@ -4,21 +4,21 @@ import {
     isTriggeredByUserInteraction,
     isPushEvent,
     isJiraWorkflowDispatchEvent,
-    isResolveConflictsWorkflowDispatchEvent, isPullRequestEvent, isPullRequestReviewEvent, isIssueCommentEvent,
+    isResolveConflictsWorkflowDispatchEvent, isPullRequestEvent, isPullRequestReviewEvent, isIssueCommentEvent, isWorkflowRunFailureEvent,
 } from "../context";
 import {checkHumanActor} from "../validation/actor";
 import {postJunieWorkingStatusComment} from "../operations/comments/feedback";
 import {initializeJunieWorkspace} from "../operations/branch";
 import {PrepareJunieOptions} from "./types/junie";
-import {detectJunieTriggerPhrase} from "../validation/trigger";
+import {detectJunieTriggerPhrase, isReviewOrCommentHasFixCITrigger} from "../validation/trigger";
 import {configureGitCredentials} from "../operations/auth";
 import {prepareMcpConfig} from "../../mcp/prepare-mcp-config";
 import {verifyRepositoryAccess} from "../validation/permissions";
 import {Octokits} from "../api/client";
 import {prepareJunieTask} from "./junie-tasks";
 import {prepareJunieCLIToken} from "./junie-token";
 import {OUTPUT_VARS} from "../../constants/environment";
-import {RESOLVE_CONFLICTS_ACTION} from "../../constants/github";
+import {RESOLVE_CONFLICTS_ACTION, FIX_CI_ACTION} from "../../constants/github";
 import {getJiraClient} from "../jira/client";
 
 /**
@@ -62,10 +62,27 @@ export async function initializeJunieExecution({
 
     // Get PR-specific info for MCP servers
     const prNumber = context.isPR ? context.entityNumber : undefined;
-    const commitSha = branchInfo.headSha;
+    
+    // For workflow_run events, use the workflow run's head_sha (the commit that was tested)
+    // This is critical for the checks server to find the correct check runs
+    let commitSha = branchInfo.headSha;
+    if (isWorkflowRunFailureEvent(context)) {
+        const workflowRunSha = (context.payload as any).workflow_run?.head_sha;
+        if (workflowRunSha) {
+            console.log(`Using workflow_run head_sha for checks: ${workflowRunSha}`);
+            commitSha = workflowRunSha;
+        }
+    }
+
+    // Detect if this is a fix-ci action (needed for auto-enabling checks server)
+    const isFixCIInPrompt = context.inputs.prompt?.includes(FIX_CI_ACTION);
+    const isFixCIInComment = isReviewOrCommentHasFixCITrigger(context);
+    const isFixCIFromWorkflowFailure = isWorkflowRunFailureEvent(context);
+    const isFixCI = isFixCIInPrompt || isFixCIInComment || isFixCIFromWorkflowFailure;
 
     // Prepare MCP configuration with automatic server activation
     // - Inline comment server: enabled for PRs (requires commitSha)
+    // - Checks server: enabled for fix-ci action or when explicitly requested
     const mcpConfig = await prepareMcpConfig({
         junieWorkingDir: context.inputs.junieWorkingDir,
         allowedMcpServers: mcpServers,
@@ -74,7 +91,8 @@ export async function initializeJunieExecution({
         repo: context.payload.repository.name,
         branchInfo: branchInfo,
         prNumber: prNumber,
-        commitSha: commitSha
+        commitSha: commitSha,
+        isFixCI: isFixCI
     })
 
     await prepareJunieTask(context, branchInfo, octokit, mcpConfig.enabledServers)
@@ -104,6 +122,10 @@ async function shouldHandle(context: JunieExecutionContext, octokit: Octokits):
         return true;
     }
 
+    if (isWorkflowRunFailureEvent(context)) {
+        return true;
+    }
+
     return isTriggeredByUserInteraction(context) && detectJunieTriggerPhrase(context) && checkHumanActor(octokit.rest, context);
 }