Tools & Skills — gitagent docs

Built-in Tools

Tool	Description	Concurrency Safe	Read Only
`cli`	Run shell commands	No	No
`read`	Read file contents	Yes	Yes
`write`	Create/write files	No	No
`memory`	Load/save persistent memory	No	No
`capture_photo`	Capture camera frame as photo	No	No
`task_tracker`	Track task progress, search skills	No	No
`skill_learner`	Save/evaluate learned skills	No	No

CLI Tool

Command: ls -la src/
Timeout: 120s (configurable)
Output: stdout + stderr (truncated to ~100KB)

Read Tool

Path: src/index.ts
Encoding: utf-8 (default) or base64
Partial reads: start/end byte offsets

Write Tool

Path: workspace/report.md
Content: "# Report\n..."
Append: false (default) — overwrites
Auto-creates parent directories

Memory Tool

load — Returns current memory/MEMORY.md content
save — Appends entry + git commits
Supports layered memory via memory.yaml
Auto-archives when max_lines exceeded (to memory/archive/<YYYY-MM>.md)

Declarative Tools (Custom)

Define tools in tools/*.yaml:

name: lookup-account
description: Look up account details by customer ID
input_schema:
  properties:
    customer_id:
      type: string
      description: The customer ID
  required: [customer_id]
implementation:
  script: scripts/lookup.sh
  runtime: sh

The script receives JSON args on stdin and outputs plain text.

Skills

Skills are reusable instruction sets that the agent follows for specific tasks.

Creating a Skill

Create skills/<skill-name>/SKILL.md:

---
name: code-review
description: Review code for bugs, style, and security issues
license: MIT
allowed-tools: cli read write
metadata:
  author: your-name
  version: "1.0.0"
  category: development
---

# Code Review

## Instructions

1. Read the specified file(s) using the read tool
2. Analyze for:
   - Bugs and logic errors
   - Security vulnerabilities (OWASP top 10)
   - Code style and readability
   - Performance issues
3. Write a review report to workspace/review.md

## Output Format

For each issue found:
- **File**: path
- **Line**: number
- **Severity**: critical / warning / info
- **Description**: what's wrong
- **Fix**: suggested change

Invoking Skills

# In REPL
/skill:code-review Review the auth module

# In voice/text
"Use the code-review skill on src/auth.ts"

Skill Learning

The agent can learn new skills automatically:

task_tracker begins tracking a task
Agent completes the task successfully
skill_learner evaluates if the approach is worth saving
If yes, crystallizes it as a new skill with confidence: 0.7
Future tasks search for matching skills
Confidence adjusts based on success/failure outcomes