AGENTS.md

1# Claude Code Configuration
2 
3This directory contains commands, agents, and skills for Claude Code.
4 
5@rules/prompt-engineering.mdc
6 
7## Structure
8 
9- `commands/` - User-invoked workflows (slash commands like `/ai-coding-config`)
10- `agents/` - Specialized AI assistants with focused expertise
11- `skills/` - Claude-invoked capabilities (activated autonomously when relevant)
12- `context.md` - Project identity and personality instructions
13 
14## Commands
15 
16Markdown files invoked via `/command-name`. Include YAML frontmatter:
17 
18```yaml
19---
20description: Brief explanation shown in command list
21argument-hint: [optional | args]
22---
23```
24 
25Commands are the canonical source - `.cursor/commands/` contains symlinks to these
26files.
27 
28## Agents
29 
30Specialized assistants with isolated context and focused tools. Frontmatter:
31 
32```yaml
33---
34name: agent-name
35description:
36  "Invoke when [trigger conditions]. Does [what it does] using [key capabilities]."
37tools: Read, Write, Edit, Grep, Glob, Bash
38model: sonnet
39---
40```
41 
42Agents run in separate context windows with only the tools listed. Use for tasks
43requiring focused expertise or isolation from main conversation.
44 
45## Skills
46 
47Capabilities Claude activates autonomously when relevant. Each skill is a directory with
48`SKILL.md`:
49 
50```yaml
51---
52name: skill-name
53description: "Use when [trigger conditions]. Does [what it does] to achieve [outcome]."
54---
55```
56 
57Skills use the main conversation context. The description is critical - Claude uses it
58to decide when to invoke the skill.
59 
60## Writing Prompts
61 
62All prompts in this directory follow prompt-engineering.mdc:
63 
64- Goal-focused instructions over step-by-step prescriptions
65- Positive patterns only - never show anti-patterns
66- XML tags for complex multi-section content
67- Trust the executing model's capabilities
68- Clarity over brevity
69