Init

Initialize the .singularskills/ directory for the current project. Set up the project configuration and choose which domains to activate.

Prerequisites

None — this is typically the first singularskills command you run in a new project.

Process

1. Check if .singularskills/ already exists:

   • If it exists, read .singularskills/config.md and show current configuration
   • Ask if the user wants to update the configuration, or only the Stack section of AGENTS.md (step 4b)
   • If it doesn't exist, proceed with initialization

2. Ask the user:

   • What is the project name?
   • Which domains will you use? Present the full list:
     • Design (component design with Figma integration)
     • Research (Double Diamond discovery and validation)
     • Development (plan, build, test, debug, review, ship)
     • Marketing (RACE Framework campaigns)
     • Branding (brand creation and compliance)
     • Accessibility (WCAG-EM audits)
     • Project Management (Scrum workflow)
     • Coaching (domain-specific mentorship)
     • Leadership (team development and management)
     • Legal (contracts, compliance, governance)

3. Create the directory structure:

   • Create .singularskills/ root directory
   • Create subdirectories for each selected domain
   • Create .singularskills/reports/ for cross-domain reports

4. Write the project configuration:

   • Write .singularskills/config.md with:
     • Project name
     • Active domains
     • Date initialized
     • Notes or goals (if the user provided any)

4b. Write the Stack section so the skills that build, test, lint, review or generate code know how this project is run. Take every value from the project's own files, never from a profile:

• stack: from package.json (packageManager, and the versions of TypeScript, Next.js, React or other frameworks among its dependencies), Cargo.toml with rust-toolchain.toml (Rust toolchain and edition), pyproject.toml, go.mod or the equivalent
• install, build, test, lint, format, typecheck, run: from package.json scripts, a justfile or a Makefile. For lint prefer the script that checks lint and format together (check, validate-style) over a bare lint; for format take only a variant that writes (format:fix, fmt, format --write) and leave format out otherwise; for test prefer the script CI runs. Leave out any key you cannot read, so the profile supplies it
• rules: from the project's README, CONTRIBUTING or conventions file (a repo-conventions skill, a CLAUDE.md rules section), as short imperatives; leave it out when the project states none
• layout: the top-level folders you see (for example apps/*, packages/*, crates/*, src/) and where tests live
• profile: bun-react-ts when package.json has packageManager starting with bun@ and both next and react among its dependencies; profile: rust-gpui when a Cargo workspace member depends on gpui; otherwise no profile key
• Keep the shape flat: every key is one line of plain text (stack, layout and rules are free text, never nested mappings or lists), and only the keys named above (stack, profile, install, build, test, lint, format, typecheck, run, layout, rules) appear; put stack first
• Write it as one fenced yaml block under a ## Stack heading. Choose the file in this order: (1) if AGENTS.md or CLAUDE.md already has a ## Stack section, replace that section in place, in that file; (2) else if CLAUDE.md exists and AGENTS.md does not, write it into CLAUDE.md; (3) else write it into AGENTS.md, creating the file with the one-line intro "This project uses singularskills; the Stack section below tells its skills how to build, test and lint it." when it does not exist. Never create AGENTS.md next to an existing CLAUDE.md just for this section, and never leave two Stack sections across the two files.

5. Suggest next steps:
   • Based on selected domains, recommend which to start with
   • Provide the first command to run

Output

The project configuration written to .singularskills/config.md and the ## Stack section written to AGENTS.md (or CLAUDE.md). Present the initialized structure and suggested next steps to the user.