Agent Skills

The repository ships three host-agnostic Agent Skills:

Skill	Purpose
orcats-setup	Install or verify the orcats CLI and verify at least one backend.
orcats-author	Generate a saved, verification-gated workflow or loop module.
orcats-flow	Run, monitor, diagnose, and heal a saved workflow or loop.

The intended sequence is:

orcats-setup -> orcats-author -> orcats-flow

Install from GitHub

orcats skills is the preferred convenience command. It requires npx from a Node.js/npm installation and delegates source retrieval, agent detection, destination, prompts, and overwrite policy to the skills CLI.

# Let the delegated installer choose skills and scope interactively.
orcats skills

# List available skills without installing.
orcats skills --list

# Install every skill globally without prompts.
orcats skills --all --global --yes

Install one skill to one agent:

orcats skills --skill orcats-setup --agent claude-code

The command supports --list, --skill <name>, --all, --agent <name>, --global, and --yes. --all and --skill cannot be combined. Without a selection option, the delegated installer stays interactive.

Direct installer fallback

Orcats does not implement agent-directory installation or package the skills in npm artifacts. The equivalent direct commands remain available:

npx skills add ASRagab/orca-ts --list
npx skills add ASRagab/orca-ts --skill orcats-setup --agent claude-code
npx --yes skills add ASRagab/orca-ts --skill '*' --global --yes

Before the repository is public, use a local checkout with the direct installer:

git clone https://github.com/ASRagab/orca-ts.git
npx skills add ./orca-ts --skill '*' --global

Each skill is self-contained. Its SKILL.md, scripts/, references, and templates are installed together.

Setup separates CLI presence, static authentication evidence, selected transport, and live readiness. It asks before the bounded readiness smoke because the smoke sends a prompt and spends tokens; declining leaves the backend unverified. Claude readiness exercises default stream-json unless the operator explicitly selected ACP.

Authoring requires offline orcats check before handoff, even when the target has no TypeScript project. Flow execution uses stderr/heartbeat, plan, loop state, and Git for live progress, then trusts terminal monitoring JSON for the final outcome. See Agent Skills Reference and Environment Variables.

Generated mutating artifacts default to baseline policy repair. Use --baseline=strict, --baseline=accept-dirty, or ORCA_BASELINE_POLICY only when you want to override that per run.