debug-pipeline
Diagnose Harness pipeline executions via MCP. Analyzes any execution (failed or successful) to produce structured reports with stage/step breakdown, timing, bottlenecks, failure details, chained pipeline drill-down, and execution logs. Use when asked to debug a pipeline, investigate a failure, find out why a build failed, analyze pipeline errors, check execution logs, review execution performance, or find bottlenecks. Trigger phrases: debug pipeline, pipeline failed, why did my build fail, analyze failure, pipeline error, execution logs, fix pipeline, execution bottleneck, slow pipeline.
Harness · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add harness/harness-skills --skill debug-pipeline技能说明
name: debug-pipeline description: >- Diagnose Harness pipeline executions via MCP. Analyzes any execution (failed or successful) to produce structured reports with stage/step breakdown, timing, bottlenecks, failure details, chained pipeline drill-down, and execution logs. Use when asked to debug a pipeline, investigate a failure, find out why a build failed, analyze pipeline errors, check execution logs, review execution performance, or find bottlenecks. Trigger phrases: debug pipeline, pipeline failed, why did my build fail, analyze failure, pipeline error, execution logs, fix pipeline, execution bottleneck, slow pipeline. metadata: author: Harness version: 2.1.0 mcp-server: harness-mcp-v2 license: Apache-2.0 compatibility: Requires Harness MCP v2 server (harness-mcp-v2)
Debug Pipeline
Diagnose pipeline executions and suggest fixes via MCP.
Instructions
Step 1: Diagnose Execution (Preferred)
Use the dedicated diagnosis tool. It accepts an execution_id, pipeline_id (auto-fetches latest execution), or a Harness URL:
Call MCP tool: harness_diagnose
Parameters:
pipeline_id: "<pipeline_identifier>" # or execution_id or url
org_id: "<organization>"
project_id: "<project>"
This returns a structured report with stage/step breakdown, timing, bottlenecks, and failure details in one call. It also automatically follows chained (child) pipeline failures.
Step 1b: Full Diagnostic Mode
For deeper analysis, request logs and pipeline YAML:
Call MCP tool: harness_diagnose
Parameters:
execution_id: "<execution_id>"
org_id: "<organization>"
project_id: "<project>"
summary: false # raw diagnostic payload
include_yaml: true # include pipeline definition
include_logs: true # include failed step logs
log_snippet_lines: 120 # tail N lines per step (0 = unlimited)
max_failed_steps: 5 # cap number of steps to fetch logs for
Diagnose Parameters
| Parameter | Default | Description |
|---|---|---|
execution_id | -- | Specific execution to analyze |
pipeline_id | -- | Fetch latest execution for this pipeline |
url | -- | Harness UI URL (auto-extracts IDs) |
summary | true | Structured report (true) or raw payload (false) |
include_yaml | false (summary) / true (raw) | Include pipeline YAML definition |
include_logs | false (summary) / true (raw) | Include failed step logs |
log_snippet_lines | 120 | Max log lines per step (tail). 0 = unlimited |
max_failed_steps | 5 | Max steps to fetch logs for. 0 = unlimited |
Step 2: Project Health Overview
Check overall project health for context:
Call MCP tool: harness_status
Parameters:
org_id: "<organization>"
project_id: "<project>"
Shows recent failed executions, running executions, and deployment activity.
Step 3: Find Failed Executions (if needed)
Call MCP tool: harness_list
Parameters:
resource_type: "execution"
org_id: "<organization>"
project_id: "<project>"
search_term: "<pipeline name>"
Step 4: Get Execution Details
Call MCP tool: harness_get
Parameters:
resource_type: "execution"
resource_id: "<execution_id>"
org_id: "<organization>"
project_id: "<project>"
Step 5: Get Execution Logs
Call MCP tool: harness_get
Parameters:
resource_type: "execution_log"
resource_id: "<execution_id>"
org_id: "<organization>"
project_id: "<project>"
Step 6: Get Pipeline Definition
Call MCP tool: harness_get
Parameters:
resource_type: "pipeline"
resource_id: "<pipeline_identifier>"
org_id: "<organization>"
project_id: "<project>"
Analysis Framework
Categorize errors and provide targeted fixes:
Build Failures
- Missing dependencies - Check package.json/requirements.txt
- Compilation errors - Review recent code changes
- Docker build failures - Check Dockerfile and base image
Infrastructure Errors
- "No delegate available" - Check delegate status, verify tags match
- Connector failures - Rotate credentials, test connection
- Resource limits - Check cloud quotas and limits
Configuration Errors
- "Secret not found" - Verify secret exists at correct scope (account/org/project)
- "Could not resolve expression" - Check expression syntax
- "Connector not found" - Verify connectorRef identifier
Deployment Errors
- ImagePullBackOff - Check registry credentials and image tag
- CrashLoopBackOff - Check container logs, resource limits
- Readiness probe failed - Review probe configuration
Timeout Errors
- Step/stage exceeded timeout - Increase timeout or optimize
- Delegate task queued too long - Scale up delegates
Artifact Errors
- "Artifact not found" - Verify artifact path, check upstream build
Response Format
## Pipeline Failure Analysis
**Pipeline:** <name>
**Execution:** <id>
**Failed At:** <timestamp>
### Failure Summary
**Stage:** <failed_stage>
**Step:** <failed_step>
**Error:** <error message>
### Root Cause
<explanation>
### Fix
**Immediate:** <specific steps>
**Prevention:** <how to avoid in future>
Examples
- "Why did my build pipeline fail?" - Use
harness_diagnosewith pipeline_id - "Debug execution abc123" - Use
harness_diagnosewith execution_id - "Show me recent failures" - Use
harness_statusthen drill into failures - "Analyze the pipeline at https://app.harness.io/..." - Pass URL directly to
harness_diagnose - "Which stage is the bottleneck in my pipeline?" - Use
harness_diagnoseon a successful execution - "Get full logs for the failed deploy step" - Use
harness_diagnosewithinclude_logs: true
Performance Notes
- Take your time analyzing logs thoroughly. Read complete error messages and stack traces before diagnosing.
- Check all failed steps, not just the first one. Multiple failures may share a root cause or reveal a dependency chain.
- Quality of diagnosis is more important than speed. A wrong diagnos