One CLI to manage skills, commands, and rules across AI coding assistants.

Your skills are scattered across ~/.claude/skills/, .cursor/rules/, CLAUDE.md, .cursorrules — different formats, different locations, no visibility into what's actually loaded. ags gives you a single command to scan, lint, measure, and manage all of it.

# Homebrew
brew install moqa-studio/tap/agents-cli

# npm / bun (requires Bun v1.0+)
bun install -g @moqa/ags

# Or download a binary from GitHub Releases

# Bash
echo 'source /path/to/agents-cli/completions/ags.bash' >> ~/.bashrc

# Zsh
cp completions/_ags /usr/local/share/zsh/site-functions/

Every command accepts --agent <name> to filter. Without it, ags operates across all agents at once.

Discover everything across all agents.

ags scan                            # all items
ags scan --agent claude             # Claude Code only
ags scan --type skill               # skills only
ags scan --scope local              # project-level only
ags scan --installed                # which agents are installed

See what's loaded into your agent's context — config files, skills, commands, memory, MCP servers.

ags context                         # all agents
ags context --agent claude          # Claude Code only

Catch issues: missing frontmatter, short descriptions, oversized files, name conflicts, unsupported keys.

ags lint                            # everything
ags lint --agent cursor             # Cursor rules only

Token budget — per-skill cost ranked by size, context usage bar per agent, suggestions to free tokens.

ags skill-cost                      # full report
ags skill-cost --scope local        # project skills only

Install a skill from GitHub.

ags grab https://github.com/org/repo/blob/main/skills/foo/SKILL.md
ags grab <url> --to cursor          # install for Cursor
ags grab <url> --dry-run            # preview only

Remove a skill, command, agent, or rule.

ags rm my-skill                     # all agents
ags rm my-skill --agent claude      # Claude Code only

Usage dashboard (Claude Code only).

ags stats                           # last 30 days
ags stats --period 7d               # last week
ags stats --period all-time         # everything

All commands support --json for structured output and --help for usage details.

bun test              # run tests
bun run typecheck     # type-check

The entire multi-agent system is driven by a single registry in src/core/agents.ts. To add an agent:

1. Add an entry to AGENTS in src/core/agents.ts — paths, binary name, config files, frontmatter keys
2. Add the name to AgentName in src/types.ts
3. Add a context limit in getContextLimit() in src/core/agents.ts
4. Add styling in AGENT_STYLE in src/utils/output.ts — icon and brand color
5. Update completions in completions/

All commands (scan, lint, skill-cost, grab, rm, context) work automatically from the registry. Agent-specific features (memory, stats, MCP) are opt-in in the relevant command file.

MIT