Skill-driven software engineering prompt library with 25 skills, 21 agents, scoped instructions, review checks, and a three-tier memory model (product plans, global knowledge, user profile). Primary platforms: GitHub Copilot in VS Code and IntelliJ IDEA.
- Clone this repository
- Open in VS Code 1.109+ with GitHub Copilot Chat
- Install harness:
npm install -g @dev-kit/harness@latest && harness installfrom Nexus (setup) — ornpm install -g ./packages/harness && harness installfrom this repo before publishing - Open a product repository (no prompt-library files copied into it); ensure
docs/plans/exists @engineer— request substantial investigation or delivery; only change-making work enters plan, implement, verify, and compound- Optional:
/harness-doctor— health check;/project-readme— README upkeep;/triage-issues— backlog triage
See Harness Quickstart.
@engineer behaves like a real engineer: starter skills, expert network, principles, compounded team knowledge, and approved growth of skills and specialists. See Engineer Harness Architecture.
Skill-first: skills are workflow contracts; agents provide isolated judgment; instructions apply by file pattern.
| Primitive | Location |
|---|---|
| Skills | .github/skills/*/SKILL.md |
| Agents | .github/agents/*.agent.md |
| Team knowledge | knowledge/ → hydrated to ~/.copilot/knowledge/ |
| Product plans | docs/plans/ in each product repo only |
Standards: Engineer Harness Architecture and Skill-Driven Prompt Library Standard.
@engineer: Answer → direct, ceremony-free reply; Investigate → evidence-backed read-only report
Deliver → orient → establish intent → investigate → work → on-demand gaps → verify → review → compound → report
open → planned → in-progress → review → done (or blocked-capability)
Pipeline steps (/capture-issue, /plan-issue, …) are engineer-internal (user-invocable: false), loaded on demand by @engineer. Optional: /brainstorming, /deepen-plan, /document-review. Plan files are the per-issue context pack. See docs/plans/_plan-template.md and capture-gate.md.
User-invocable: /engineer, /harness-doctor, /project-readme, /triage-issues. All other skills are engineer-internal, loaded on demand by @engineer.
| Skill | Type | Access | Purpose |
|---|---|---|---|
/engineer |
Engineering | User | Accountable full-cycle coordinator |
/harness-doctor |
Utility | User | Hydrate and harness health check |
/project-readme |
Documentation | User | README maintenance |
/triage-issues |
Utility | User | Backlog prioritization |
/recall |
Memory | Internal | Team + repo knowledge before work |
/index-memory |
Memory | Internal | Rebuild knowledge/manifest.yaml |
/consolidate |
Memory | Internal | Convert unconsolidated episodes into learnings via harness consolidate |
/capture-issue |
Pipeline | Internal | Create product plan file |
/plan-issue |
Pipeline | Internal | Research and lock plan |
/code-review |
Pipeline | Internal | Confidence-scored review |
/compound-learnings |
Pipeline | Internal | Publish to global knowledge/solutions/ |
/brainstorming |
Extension | Internal | Requirements exploration |
/deepen-plan |
Extension | Internal | Interactive plan deepening |
/document-review |
Extension | Internal | Document quality gate |
/create-primitive |
Extension | Internal | Approved primitive creation |
/import-conventions |
Extension | Internal | Import external conventions |
/java, /python, /sql, /aws |
Domain | Internal | Domain workflows |
/codebase-context |
Utility | Internal | Architecture snapshot |
/ensure-plan, /ensure-capability, /auto-compound, /auto-skill-draft |
Internal | Internal | Planning, on-demand gap resolution, automatic post-success learning, and experimental skill drafting |
17 specialists + @engineer + @code-implementer + 2 coordinators (plan-coordinator, code-review-coordinator). Only @engineer is user-invocable; coordinators are internal, dispatched by @engineer. Inventory: CLAUDE.md or knowledge/capability-registry.yaml.
| Tier | Where | What |
|---|---|---|
| Team (global) | knowledge/solutions/ + manifest.yaml |
Cross-repo learnings after hydrate |
| Product | docs/plans/ |
Active issues (local only) |
| Product (optional) | docs/solutions/ |
Repo-private learnings |
| User | ~/.copilot/knowledge/profile.md |
Preferences |
Context lookup order: .github/skills/references/knowledge-locations.md.
.github/ agents, skills, instructions, copilot-instructions.md
knowledge/ team solutions, manifest, capability-registry (hydrated globally)
docs/architecture/ canonical harness architecture and primitive standard
docs/plans/ template + at most one live PR plan (see docs/plans/README.md)
.vscode/ @dev-kit/harness tasks (install/upgrade/doctor), MCP config
AGENTS.md Cross-tool guidance
- VS Code 1.109+ with GitHub Copilot Chat (or current IntelliJ Copilot with global customizations)
- Global install via
@dev-kit/harness(see Install)