cursor-jobs-overview.md

open-agents — the delegation primitive

The canonical source of the collaboration primitive — the open-agents core + the explore/fast-edit skills that let the orchestrator delegate scoped jobs to cursor-agent -p and self-wait on them.

• explore/skill/SKILL.md, fast-edit/skill/SKILL.md — the skill definitions Claude Code loads (explore = read-only quote-extractor; fast-edit = one-pass mutator); each ships a bundled scripts/<name>.ts
• open-agents/src/ — cli.ts + core/ · profiles/ · adapters/ · entrypoints/ (the self-waiting job runner, Bun) — the shared, editable source the skill scripts are built from
• npx skills add installs the skills into ~/.claude/skills
• experiments/ — strategy-arm runs; per-skill feedback/ — lived-use friction logs; notes/ — the explore/mechanical-edit doctrine

Usage in one line (see explore/skill/SKILL.md for the full contract): write a scoped spec.md, then

bun ~/.claude/skills/explore/scripts/explore.ts submit --file spec.md

it blocks, writes each job's clean deliverable to its own file, and you verify the load-bearing fact yourself.

Install model

npx skills add DawidWraga/skills installs the three skills into ~/.claude/skills/{explore,fast-edit,diagnose} (install identity is the SKILL.md name: frontmatter, not the folder name). Each skill ships a self-contained, dependency-free scripts/<name>.ts built from open-agents/src/ via bun run build — no shim, no junctions, no post-install build hook.

open-agents/ has no SKILL.md, so it is never installed as a skill — it is the shared, editable source the committed skill scripts are generated from. Edit open-agents/, run bun run build, never hand-edit scripts/.

The measurement scripts the experiments use live in open-agents/observability/ (Bun+TS, single runtime), shared across all skills in this repo.