CLI Reference¶

The reach command-line interface provides tools for the complete skill measurement and optimization lifecycle.

Command Overview¶

Command	Purpose
`reach check`	Execute two-stage CI/CD quality gate combining linting and empirical assertions.
`reach clean`	Clean cached Agent Registry payloads, sandboxes, and evaluation artifacts.
`reach cluster`	Partition skill catalogs into cohesive subagent scopes to prevent routing decay.
`reach completion`	Generate or install shell tab completion scripts for bash, zsh, and fish.
`reach diff`	Change one factor (description, rival, scope), hold queries fixed, and report the delta.
`reach doctor`	Inspect and diagnose local environment, agent binaries, credentials, and catalogs.
`reach eval`	Measure whether a catalog's skills are reachable, writing a query set if none exists.
`reach init`	Scaffold reach.toml configuration and initialize skill directories.
`reach lint`	Validate skill manifests, frontmatter schemas, naming, and runtime limits.
`reach optimize`	Optimize a skill's description using candidate synthesis and empirical probes.
`reach overlap`	Find which of your installed skills compete to answer the same requests.
`reach query`	Draft, export, import, and view query datasets.
`reach sweep`	Execute multi-scale catalog evaluation sweeps to measure reachability decay.
`reach view`	Read back a recorded run and render what it measured (terminal or HTML).

The following flags apply across reach commands:

--help, -h: Display command help and exit.
--version: Display application version and exit.
--global, -g: Discover and inspect skills from user global configuration (~/.agents/skills, ~/.claude/skills, ~/.cursor/skills, etc.). Can be used as a top-level flag (reach --global lint) or subcommand flag (reach lint --global).
--quiet, -q: Suppress interactive progress bars and spinners.
--verbose: Display full cryptographic hashes and detailed logs.

When running commands that interact with an agent runtime (check, eval, optimize), select the target runtime via --agent <runtime>:

claude-code: Drives Anthropic's Claude Code CLI.
antigravity-cli: Drives the Google Antigravity Agent API / CLI.
antigravity-sdk: Uses the Google Antigravity Python SDK directly.
goose: Drives the Goose agent CLI (aaif-goose/goose).
pi: Drives the headless Pi coding agent CLI (earendil-works/pi).
keyword: Fast rule-based lexical matching runtime for instant baseline scores.