Our review

This skill delegates tasks to specialized GLM-5 agents with thinking mode enabled, enabling parallel work and code review.

Strengths

Delegation to agents with predefined roles (coder, reviewer, tester) for efficient specialization.
Built-in thinking mode that enhances the quality of agent responses.
Integration with existing commands (orchestrator, parallel, loop) for complex workflows.

Limitations

Requires access to the GLM-5 API, which may be costly or limited in availability.
Generates log and status files in the .ralph directory, adding noise to the project.
Agents are not persistent; each invocation creates a new context.

When to use it

Use this skill when you need specialized subtasks (coding, review, tests) to run in parallel or sequentially with deep reasoning.

When not to use it

Avoid using it for simple, quick tasks where a single agent without thinking is sufficient, or when the GLM-5 API is unavailable.

VERSION: 2.87.0

name: glm5 description: GLM-5 Agent Teams skill for spawning teammates with thinking mode allowed-tools:

Bash
Read
Write

GLM-5 Agent Teams Skill

Spawn GLM-5 powered teammates with native thinking mode.

Quick Usage

# Spawn single teammate
/glm5 coder "Implement auth"

# Spawn for review
/glm5 reviewer "Review this code"

# Spawn for testing
/glm5 tester "Generate tests"

Integration with Other Commands

This skill integrates with:

/orchestrator - GLM-5 as teammate option
/parallel - GLM-5 for parallel review
/loop - GLM-5 for iterative tasks

Agent Types

| Type | Role | Best For | |------|------|----------| | coder | Implementation | Features, refactoring, bugs | | reviewer | Code Review | Security, quality, patterns | | tester | Test Generation | Unit tests, coverage | | planner | Architecture | Design, planning | | researcher | Documentation | Docs, exploration |

Execution

When this skill is invoked:

Parse Arguments: Extract role and task from $ARGUMENTS
Generate Task ID: task-{timestamp}
Call GLM-5 API: With thinking mode enabled
Capture Output: Reasoning + result
Fire Hooks: SubagentStop (native Claude Code hook)

Bash Commands

Spawn Teammate

.claude/scripts/glm5-teammate.sh <role> "<task>" "<task_id>"

Check Status

cat .ralph/team-status.json

View Logs

tail -f .ralph/logs/teammates.log

Output Files

| File | Content | |------|---------| | .ralph/teammates/{id}/status.json | Task status & metadata | | .ralph/reasoning/{id}.txt | GLM-5 reasoning | | .ralph/logs/teammates.log | Activity log |

Example Session

User: /glm5 coder "Implement factorial in TypeScript"

[GLM-5 thinking...]
Reasoning: The user wants a TypeScript factorial function...
Output: function factorial(n: number): number { ... }

✅ Task completed
📁 Status: .ralph/teammates/task-123/status.json
🧠 Reasoning: .ralph/reasoning/task-123.txt

GLM-5 Agent Teams with Thinking Mode

Recommended for