name: 17th-make-triager description: | ALWAYS use this skill when user requests any make command: make test, make lint, make build, make clean, make test-unit, make test-integration, make test-prepare, or any other Makefile target. This skill delegates make execution to GLM worker which runs the command, analyzes output, and creates a report file. Do NOT use for: hooks, automated checks, file-change triggered linting. Requires glm-worker MCP to be configured.

Make Triager

Delegate make commands to GLM-5 worker for execution and intelligent failure triage.

Prerequisites

Configure glm-worker MCP in Claude Code's mcp.json:

{
  "mcpServers": {
    "glm-worker": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/glm-worker-mcp", "python", "server.py"],
      "env": {
        "ZHIPU_API_KEY": "your-api-key"
      }
    }
  }
}

Workflow

1. Receive make command request from user
2. Get current git branch name
3. Call delegate_to_glm with branch name and triager prompt template
4. GLM executes command and analyzes failures if any
5. GLM writes failure report to .kkachi/{branch_name}/completed/
6. Return report path to main agent for detailed analysis

Prompt Template

Use delegate_to_glm with this template. Replace {branch_name} with current git branch:

delegate_to_glm(
    task="""
Execute: {make_command}

Always write report to: .kkachi/{branch_name}/completed/make-triager-{epoch_timestamp}.md
- Use current Unix epoch timestamp in seconds for {epoch_timestamp}

On success:
- Write success report
- Return "✅ {make_command} passed. Report: {report_path}"

On failure:
- Extract ALL failed items (DO NOT truncate - list every single failure)
- Group failures by root cause/error pattern
- Identify if multiple failures share the same underlying issue
- Write comprehensive failure report
- Return "❌ {make_command} failed. {N} failures. Report: {report_path}"

Success report format:
# Success: {command}

**Timestamp**: {iso_timestamp}
**Exit Code**: 0

✅ All checks passed.

Failure report format:
# Failure Report: {command}

**Timestamp**: {iso_timestamp}
**Exit Code**: {exit_code}
**Total Failures**: {count}

## Summary
(1-2 sentence overview: what failed, common patterns if any)

## Failures by Root Cause

### Root Cause 1: {description}
Affected: {N} items
- item1 (file:line)
- item2 (file:line)

Error pattern:

{common error message}

### Root Cause 2: {description}
(repeat for each distinct root cause)

## All Failed Items (Complete List)
1. item1 (file:line) - [Root Cause 1]
2. item2 (file:line) - [Root Cause 1]
3. item3 (file:line) - [Root Cause 2]
(list ALL failures, no truncation)

## Suggested Actions
1. Fix for Root Cause 1: {specific action}
2. Fix for Root Cause 2: {specific action}
""",
    working_dir="{project_root}"
)

Variables

Examples

Single Command

delegate_to_glm(
    task="""
Execute: make lint

Always write report to .kkachi/feature/task-177/completed/make-triager-{epoch}.md

On success: Return "✅ make lint passed. Report: <path>"
On failure: Extract errors with file:line, return "❌ make lint failed. Report: <path>"
""",
    working_dir="/home/user/project"
)

Multiple Commands

delegate_to_glm(
    task="""
Execute in order, stop on first failure:
1. make lint
2. make test-unit
3. make test-integration

Write report to .kkachi/feature/task-177/completed/make-triager-{epoch}.md

Return:
- All passed: "✅ All checks passed. Report: {path}"
- Any failed: "❌ {failed_cmd} failed. Report: {path}"
""",
    working_dir="/home/user/project"
)

Post-Triage Actions

After receiving a report path from GLM:

1. Read the report file
2. If success: acknowledge and proceed
3. If failure:
   • Analyze failures in context of the codebase
   • Propose specific fixes with code changes
   • Optionally re-run failed tests after fixes

Output Directory

Reports saved to .kkachi/{branch_name}/completed/ in project root. Add to .gitignore:

.kkachi/