Agent Tools Reference

This skill covers the tools available to Overcut agents, built-in agent type presets, and the persistent memory system.

Tool Categories Overview

Overcut provides 35 user-configurable tools plus 6 system tools, organized by category:

Built-in Agent Types

When creating agents in Overcut, you select a base type that determines the default tool set:

Key differences:

• •SeniorDeveloper can create PRs, merge PRs, delete files — full write access
• •CodeReview cannot create PRs or delete files, but can submit reviews and append to files
• •TechWriter has nearly the same tools as SeniorDeveloper (full write access for docs)
• •ProductManager cannot create PRs, write files, or delete files — primarily read + ticket management
• •ExploreAgent is minimal — read-only code exploration only

Automatic Tool Injection

Regardless of base type, these tools are automatically added at runtime:

• •Memory tools (memory_write, memory_read) — added to all agents (coordinator and sub-agents)
• •Coordinator tools (delegate_to_sub_agent, update_status, task_completed) — added only to coordinator agents in agent.session steps

Memory System

The memory system allows agents to persist and retrieve learnings across workflow executions.

memory_write

{
  "title": "Short summary of the learning (1 line)",
  "content": "Full details of the learning"
}

• •title: Short summary used as the identifier (1 line)
• •content: Full details of the learning

memory_read

{
  "title": "Memory title to retrieve"
}

• •title: Case-insensitive match against stored memory titles

Memory Behavior

• •Memories persist across workflow executions for the same agent
• •Auto-injected into agent context at the start of each execution
• •Use for: repository conventions, past review patterns, common issues, team preferences
• •Keep titles descriptive and unique for reliable retrieval

Tool Selection for Prompts

When writing step prompts, restrict tools to only what's needed. Use allowed/prohibited tool tables:

### Allowed Tools
| Tool | Purpose |
|------|---------|
| `read_file` | Read source files for analysis |
| `code_search` | Find relevant code patterns |
| `add_comment_to_pull_request` | Post review summary |

### Prohibited Tools
❌ `write_file` — this is a read-only analysis step
❌ `run_terminal_cmd` — no terminal commands needed
❌ `create_pull_request` — PR already exists

This pattern appears in many real playbooks and significantly improves agent focus and efficiency.

For the complete tool listing with tool identifiers, see references/tool-catalog.md.