---

name: inspecting-codefresh-failures description: Instruction guide for analyzing Codefresh CI failures. When invoked, IMMEDIATELY follow the step-by-step instructions to extract build IDs, fetch logs, and create a formatted failure report. Use when reviewing recursionpharma PRs with failing CI. allowed-tools: [Bash, Read]

Inspecting Codefresh Failures

⚡ THIS IS AN INSTRUCTION GUIDE - EXECUTE IMMEDIATELY

After invoking this skill, DO NOT WAIT - immediately execute the following steps using the PR data you already have:

QUICK START - Execute Now:

STEP 1: Extract build ID from the failing check

• Look for "state": "FAILURE" or "conclusion": "FAILURE" in statusCheckRollup
• Find the Codefresh URL in targetUrl or detailsUrl
• Extract the build ID (24-char hex) from https://g.codefresh.io/build/{build-id}

STEP 2: Fetch build metadata

Copycodefresh get builds <build-id> -o json

STEP 3: Search logs for errors

Copycodefresh logs <build-id> | grep -E -B 10 -A 20 -i "error|fail|found.*error"

STEP 4: Format your findings using the "Expected Output Format" template below

STEP 5: Include the formatted report in your PR review under "## CI Failure Analysis"

---

Detailed Instructions

This skill provides step-by-step guidance for analyzing Codefresh CI failures in recursionpharma organization repositories.

Prerequisites

Required:

• codefresh CLI must be installed (typically via Homebrew: brew install codefresh)
• User must be authenticated with Codefresh (check with codefresh auth current-context)

If prerequisites are missing:

• Inform the user that the Codefresh CLI is not available
• Suggest they install it: brew install codefresh
• Provide alternative: manually check the Codefresh build URL in the browser

When to Use This Skill

Use this skill when:

• You see a failing CI check in a recursionpharma PR
• User mentions CI failures in recursionpharma repos
• Debugging "passes locally but fails in CI" issues
• Investigating build failures from Codefresh

What This Skill Does

1. Accepts input:

   • Build ID directly (e.g., 67c7469b3275b6f1b9f96f69)
   • PR context with statusCheckRollup containing Codefresh URL
   • Repository and PR number (will fetch status checks)

2. Analyzes the failure:

   • Fetches build metadata via codefresh get builds <build-id> -o json
   • Retrieves full logs via codefresh logs <build-id>
   • Searches for error patterns in logs
   • Identifies failed step and specific error messages

3. Returns structured report with:

   • Build metadata (pipeline, duration, status, branch)
   • Failed step name
   • Specific error messages with file:line references
   • Root cause analysis
   • Impact assessment (critical vs non-blocking)
   • Whether errors are from PR changes or pre-existing

Usage Examples

Example 1: From Build ID

Copy# If you already have the build ID
Build ID: 67c7469b3275b6f1b9f96f69

Example 2: From PR Status Check Data

Copy{
  "statusCheckRollup": [
    {
      "__typename": "StatusContext",
      "context": "python/library",
      "state": "FAILURE",
      "targetUrl": "https://g.codefresh.io/build/67c7469b3275b6f1b9f96f69"
    }
  ]
}

Example 3: From Repository and PR Number

Copy# Will fetch status checks and extract build ID
Repository: recursionpharma/phenomics-potency-prediction
PR Number: 2

How to Extract Build ID

From PR JSON data (gh pr view --json statusCheckRollup):

1. Look for failing checks:

   • "state": "FAILURE" (StatusContext)
   • "conclusion": "FAILURE" (CheckRun)

2. Find Codefresh URLs in:

   • targetUrl field (StatusContext)
   • detailsUrl field (CheckRun)

3. Extract build ID from URL:

   • Format: https://g.codefresh.io/build/{build-id}
   • Example: 67c7469b3275b6f1b9f96f69

Commands Used

Copy# Get build metadata
codefresh get builds <build-id> -o json

# Get full logs
codefresh logs <build-id>

# Search for errors (most efficient - use multiple patterns)
codefresh logs <build-id> | grep -E -B 10 -A 20 -i "error\|fail\|found.*error"

# Common error patterns to look for:
# - "Found N errors" (linting/type checking)
# - "FAILED" or "FAILURE" (test results)
# - "Error:" or "ERROR:" (build errors)
# - "exit code" (non-zero exit codes)

Error Handling

If the codefresh command is not available:

Copy# Check if codefresh is installed
which codefresh || echo "Codefresh CLI not installed"

# If not installed, inform user and provide fallback:
# "The codefresh CLI is not installed. You can:
# 1. Install it: brew install codefresh
# 2. View the build in browser: [build URL from status check]"

Expected Output Format

Returns markdown-formatted analysis ready to include in PR reviews or bug reports:

Copy## CI Failure Analysis

**Build**: python/library (failed after 5m 2s)
**Failed step**: Check Code with Ruff
**Branch**: legion-2025-03-04-f8ef390a
**Commit**: abc1234...

### Errors Found (10 total)

1. **`phenomics_potency_prediction/utils.py:331,333,365,367,369`**
   - Undefined name `random_seed` (5 occurrences)
   - **Impact**: Likely runtime errors - CRITICAL to fix
   - **Type**: F821 (undefined name)

2. **`phenomics_potency_prediction/gene-cossim-curvefit/parallel-curvefit.py:233`**
   - Line too long (3373 characters > 120 limit)
   - **Impact**: Code quality issue - non-blocking
   - **Type**: E501 (line too long)

3. **`parallel-curvefit.py:139,275`**
   - Bare `except` clauses (2 occurrences)
   - **Impact**: Poor error handling - should specify exception types
   - **Type**: E722 (bare except)

### Root Cause

Pre-existing code quality issues exposed by recent infrastructure changes (such as updates to dependency sources). The changes only modify dependency sources in files like `Dockerfile.test` and `tox.ini`, not application code.

### Recommendation

Request changes - fix the 5 undefined `random_seed` errors before merging (critical runtime bugs). Other linting issues can be addressed in a follow-up PR.

Notes

• Only works for recursionpharma org repos - Codefresh is their CI system
• Build IDs are unique - Each build has a unique 24-character hex ID
• Logs can be large - Use grep patterns to find relevant errors quickly (logs can exceed 30k lines)
• Context matters - Always distinguish PR-introduced errors from pre-existing issues
• Tested and verified - Commands tested against build 67c7469b3275b6f1b9f96f69 (recursionpharma/phenomics-potency-prediction#2)

Common error patterns:

• Linting errors (Ruff, Flake8, Black) - Look for "Found N errors"
• Type errors (mypy, pyright) - Look for "error:" in output
• Test failures (pytest, unittest) - Look for "FAILED" or "FAILURE"
• Build failures (missing dependencies, import errors) - Look for "Error:" or "exit code"
• Security scans (dependency vulnerabilities) - Check specific security step logs

---

🔴 IMPORTANT REMINDERS

You are reading this skill because you are reviewing a PR with failing CI.

RIGHT NOW, you should:

1. ✅ Have already fetched the PR data with gh pr view <number> --json statusCheckRollup
2. ✅ See failing checks in that data with "state": "FAILURE" or "conclusion": "FAILURE"
3. ⚠️ NEXT: Extract the build ID from the Codefresh URL and run the commands above
4. ⚠️ DO NOT: Wait, ask questions, or pause - execute the steps immediately
5. ⚠️ GOAL: Create a "## CI Failure Analysis" section in your PR review with specific errors

If you find yourself waiting or unsure what to do next, re-read the QUICK START section at the top.