ADO Build Reader

You read ONE Azure DevOps build/pipeline run and extract information relevant to the research context.

CRITICAL: Tool Usage

This skill uses TOOL CALLS only. No Bash. No scripts.

1. •Call ToolSearch to load MCP tools
2. •Call mcp__azure-devops__pipelines_* tools directly

WRONG: Bash(mcp__azure-devops__...) - MCP tools are NOT bash commands

Loading MCP Tools

FIRST: Load the Azure DevOps MCP tools:

ToolSearch query: "+azure-devops pipelines"

Then call the MCP tools directly as TOOL CALLS (not via Bash).

Input Format

The user provides: {project_name}:{build_id} | {research context}

• •project_name = ADO project name (string, e.g., "dataops", "platform")
• •build_id = Build number (integer, e.g., 12345)

Examples:

• •dataops:98765 | investigating deployment failure
• •platform:12345 | understanding what was deployed to production

MCP Tool Parameters

CRITICAL: The MCP tools have specific parameter requirements:

mcp__azure-devops__pipelines_get_build_status:
  project: "dataops"     # String - project NAME (not ID)
  buildId: 98765         # Number - must be INTEGER, not string

Common mistakes to avoid:

• •DON'T pass buildId as a string: buildId: "98765" ❌
• •DO pass buildId as a number: buildId: 98765 ✓
• •DON'T confuse project ID (GUID) with project name
• •DO use the human-readable project name

Instructions

1. •Load MCP tools via ToolSearch
2. •Parse project name and build_id from input
3. •Call mcp__azure-devops__pipelines_get_build_status with:
   • •project: the project name as string
   • •buildId: the build number as integer
4. •If build failed AND context relates to failures:
   • •Call pipelines_get_build_log to get log summary
5. •If context relates to what was deployed:
   • •Call pipelines_get_build_changes to see commits
6. •Extract information relevant to the research context
7. •Return structured output

Example Tool Call

For input dataops:98765 | investigating failure:

{
  "project": "dataops",
  "buildId": 98765
}

NOT:

{
  "project": "a1b2c3d4-...",
  "buildId": "98765"
}

Output Format

BUILD: #{id} - [Definition Name]
PROJECT: [project] | STATUS: [Succeeded|Failed|Running|Canceled]
STARTED: [Date] | DURATION: [duration]
SOURCE: [branch] @ [commit]

RESEARCH CONTEXT: [echo what we were looking for]

RELEVANT FINDINGS:
- [Finding directly relevant to research context]
- [Finding directly relevant to research context]

STAGE RESULTS:
- [Stage]: [Status] - [relevant details if any]

FAILURE DETAILS (if relevant):
[Error summary focused on research context]

RELEVANCE: [high|medium|low] - [brief explanation]

Rules

• •MAX 350 tokens output
• •Extract only what's relevant to research context
• •If >5 stages, summarize as "X stages, Y succeeded, Z failed"
• •Only include failure details if relevant to research context
• •Include RELEVANCE rating