@respec-gtc/cli

CLI for managing shared AI coding skills via Respec.

Install

npm install -g @respec-gtc/cli

Core trust workflow

For repo-native Respec Core adoption, the main pipeline is:

respec lint <repo>
respec migrate <repo>
respec activate <repo>
respec doctor <repo>

Mental model:

• lint = trust floor
• migrate = smallest honest rewrite toward the current contract
• activate = bind the repo to the runtime
• doctor = reality check after binding

Commands

| Command | Description | |---------|-------------| | respec init | Initialize Respec in a project | | respec pull | Pull skills from remote GitHub repo | | respec push | Push local changes to your user branch | | respec export | Generate rules files from local skills | | respec status | Show local vs remote diff summary | | respec lint | Validate a Respec Core pack or standalone skill tree | | respec migrate | Preview or apply safe legacy Core-manifest rewrites | | respec activate | Write repo-local activation metadata or approved runtime bridge changes | | respec doctor | Diagnose repo/runtime/env issues after lint and activation |

Quick Start

Project initialization

npm install -g @respec-gtc/cli
respec init

Set GITHUB_TOKEN or RESPEC_GITHUB_TOKEN for pull/push/status commands.

Core package workflow

# Inspect current repo/pack state
respec lint /path/to/repo

# Preview legacy-field rewrites without touching files
respec migrate /path/to/repo

# Apply only safe manifest rewrites
respec migrate --write /path/to/repo

# Bind the repo locally
respec activate /path/to/repo

# Confirm reality matches intent
respec doctor /path/to/repo

respec migrate behavior

respec migrate is intentionally conservative.

Safe rewrites currently include:

• activation.skillPath -> activation.skill_dir
• add missing activation.assistant_contract
• synthesize missing compatibility fields from contract_version

If the manifest contains risky semantic mismatches — for example an impossible compatibility window or contract claims that contradict each other — the command stops at review-required instead of inventing new truth.