Features
A detailed overview of everything Agent Tools Loadout can do.
Content Browser
The sidebar tree view organizes all discovered content by source repository and content type.
Tree structure
my-team-rules (12 items)
├── Instructions
│ ├── ☑ Clean Code Standards
│ ├── ☐ API Design Guidelines
│ └── ☐ Testing Requirements
├── Skills
│ ├── ☐ PR Security Review
│ └── ☐ Database Migration Helper
└── Sub-agents
└── ☐ Code Review Expert
community-prompts (5 items)
└── Skills
├── ☐ React Component Generator
└── ☐ Git Commit Writer
- Source nodes show the repository name and item count
- Type groups appear when a source has multiple content types
- Content items have checkboxes for batch selection
- Items display author name and relative date when git metadata is available
Content type icons
| Icon | Type | Description |
|---|---|---|
:material-book-open-variant: book |
Instructions | Rules, guidelines, coding standards |
:material-auto-fix: wand |
Skills | Commands, prompts, specialized capabilities |
:material-account: person |
Sub-agents | Agent personas, delegated roles |
Rich Preview Panel
Click any content item to open a preview panel showing:
- Name and description from front-matter metadata
- File path in the source repository
- Metadata badges — content type, tags, tech stack, difficulty level
- Author and version if specified in front-matter
- Git history — last author, date, and commit message
- Full content with syntax highlighting
The preview panel reuses the same tab for each file, keeping your editor clean.
Equipped State Tracking
After equipping content, the extension tracks what's loaded in your workspace:
Loaded indicator
Equipped items show:
- A blue checkmark icon replacing the content type icon
- (loaded) label in the item description
- Context value changes to content-loaded
Stale detection
When the source repository is updated after you equipped an item:
- A warning icon replaces the checkmark
- Description changes to updated since loaded
- This means the source file has a newer git date than when you loaded it
To resolve a stale item, simply re-equip it — the file is overwritten with the latest version and the record is updated.
Pruning
On each refresh, the extension checks if equipped files still exist on disk. If a file was manually deleted, its tracking record is automatically removed.
Search
Search across all loaded repositories using the search button in the sidebar title bar.
How search works
- Enter one or more search terms separated by spaces
- All terms must match (AND logic)
- Search is case-insensitive
- Results update with a 250ms debounce
Searchable fields
| Field | Example |
|---|---|
| Name | "Clean Code Standards" |
| Description | "guidelines for writing clean code" |
| Content type | "instructions", "skill", "subagent" |
| File path | "rules/typescript" |
| Author | "alice" |
| Tags | "react", "testing" |
| Tech stack | "typescript", "nextjs" |
| Level | "advanced", "beginner" |
Search example
Searching for react typescript will match items that contain both "react" and "typescript" in any combination of the searchable fields.
Clearing search
Click the X button in the sidebar title bar, or run Agent Tools Loadout: Clear Search from the Command Palette.
Type Filter
Filter the tree to show only one content type at a time.
- Click the filter icon in the sidebar title bar
- Select: Instructions, Skills, or Sub-agents
- The tree updates to show only matching items
The filter icon changes to filled when a filter is active. Click it again to clear.
Note
Search and type filters can be combined. For example, filter to "Skills" and search for "react" to find React-specific skills across all sources.
Cross-Agent Format Conversion
Equip conversion
When you equip content, the extension automatically converts it to the target agent's native format:
- Frontmatter is rewritten for the target agent
- File extension matches the agent's convention
- Directory placement follows the agent's expected structure
See Supported Agents for the full format specification.
Standalone conversion
The Convert command converts an existing agent configuration file to another format:
- Open the Command Palette
- Run Agent Tools Loadout: Convert
- Select the source agent (the format you're converting from)
- Select the target agent (the format you want)
The source file is read, normalized, and written in the target format. Both files remain in your workspace.
Example conversions:
- .cursorrules → .github/copilot-instructions.md
- .github/copilot-instructions.md → CLAUDE.md
- CLAUDE.md → .cursorrules
Content Scoring
The extension uses a point-based scoring system to determine which files are likely instruction content. This happens automatically during scanning.
Score breakdown
| Signal | Points | Description |
|---|---|---|
| Known filename | +30 | Files like .cursorrules, claude.md, copilot-instructions.md |
| Inside instruction directory | +30 | Files in rules/, skills/, agents/, .github/, .claude/, etc. |
| Filename pattern | +20 | Extensions like .agent.md, .prompt.md, .mdc |
Front-matter type field |
+40 | Strongest signal — explicit type declaration |
Front-matter description field |
+10 | Indicates structured metadata |
| Content heuristics (2+ matches) | +20 | Patterns like "you are a", "## instructions", "your role is" |
| Content heuristics (1 match) | +10 | Single pattern match |
Maximum score: 100 points
Sensitivity thresholds
The sensitivity setting determines the minimum score required for a file to appear in the browser:
| Level | Threshold | What gets through |
|---|---|---|
| Low | 10 points | Almost everything — any file with a matching extension in the repo |
| Medium | 40 points | Files with at least one structural signal (directory, frontmatter, or pattern) |
| High | 70 points | Only files with strong signals (explicit type + content patterns) |
Caching and Performance
Shallow clones
Repositories are cloned with --depth=1 to minimize download size and time.
Worker thread
Content scanning runs in a separate worker thread so the VS Code UI stays responsive during large scans.
Scan caching
After the initial scan, subsequent refreshes skip re-scanning repos where HEAD hasn't changed. Use Purge Cache to force a full re-clone.
Parallel sync
All configured repositories are cloned and pulled concurrently.
Single git log
Git metadata (author, date, commit message) is collected with a single git log call per repo, avoiding N+1 query patterns.