feat: generate skills to workspace root with composable naming and YAML frontmatter

[pyramation]

feat: generate skills to workspace root with composable naming

Summary

Refactors codegen skill generation to output skills to the workspace root skills/ directory (instead of nested inside each generator’s output like orm/skills/, hooks/skills/, cli/skills/), using Agent Skills format (SKILL.md + YAML frontmatter) and composable naming to avoid collisions across targets and generator types.

Skills generation is controlled via docs.skills (default false).

Key changes:

• New workspace.ts: findWorkspaceRoot() walks up from the output dir looking for pnpm-workspace.yaml → lerna.json → package.json with workspaces → nearest package.json
• Config:
  • docs.skills?: boolean (default false) gates skill generation
  • Optional skillsPath?: string on GraphQLSDKConfigTarget to override the output directory; otherwise uses {workspaceRoot}/skills/
• Composable naming (collision-proof across generator types + targets):
  • ORM: orm-{target}-{entity}/SKILL.md
  • Hooks: hooks-{target}-{entity}/SKILL.md
  • CLI: cli-{target}-{entity}/SKILL.md (plus cli-context/ and cli-auth/ for infra)
• YAML frontmatter: all skill files include:

  ---
  name: ...
  description: ...
  ---

Updates since last revision

• Removed automatic deletion of any “old nested” skill directories (orm/skills, hooks/skills, cli/skills). No migration/cleanup behavior is performed by the generator.

Review & Testing Checklist for Human

• CLI naming collision risk: table skills and custom operation skills both use cli-{target}-{kebab}. If a table and operation share the same kebab-cased name, one will overwrite the other. Consider disambiguating (e.g. cli-{target}-table-{kebab} vs cli-{target}-op-{kebab}).
• docs.skills toggle behavior: confirm skills are only generated when docs.skills: true and default behavior (omitted) does not emit skills.
• Single-target targetName defaults to 'default': when generate() is called directly (not via generateMulti), skills are named *-default-{entity}. Confirm this is desired vs inferring from config.
• No stale skill pruning: skills are written with pruneStaleFiles: false (removed entities leave orphan skill dirs). Confirm this is intentional.
• Run real SDK generation (e.g. pnpm generate:sdk) and confirm:
  • skills land in {workspaceRoot}/skills/ (or skillsPath), not under src/**
  • YAML frontmatter + naming conventions match expectations
  • if you previously generated nested skill dirs, you may need to delete those manually

Recommended test plan: run the SDK generation script(s) that your release process uses and inspect the resulting {workspaceRoot}/skills/ directory for naming, frontmatter, and collisions.

Notes

• pnpm build passes at repo root
• All graphql/codegen tests pass
• Link to Devin Session: https://app.devin.ai/sessions/56210b5075d14c03abb8d475e781b23e
• Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring