The skill pack behind the Engineer archetype (engineer-archetype): a hands-on pair-programming navigator. The operator drives and authors every change; the agent orients, reproduces, points at evidence, and reviews. Prompt-only: no tools, routes, surfaces, or config of its own.
| Skill | What it does |
|---|---|
repo-onboard |
Clone or register a repo, detect its ecosystem(s) and show the operator the README (a markdown artifact) and the primary manifest (code pane), draw a 5–12-node architecture overview as a Mermaid flowchart whose every node links to anchored code, prove the toolchain (mise-aware), take a baseline, write a ≤25-line repo card (saved to memory), then a guided tour of a key flow, one stop per turn. |
debug-loop |
Seven guided steps, one checkpoint per turn: orient → reproduce → narrow → their hypothesis → locate (the operator types the fix) → review + verify → the operator commits. Ends with an appendix of why LLM/agent apps "stop without answering". |
Both are agent-retrievable (not slash-only): the Engineer persona reaches for them on
its own. They point at code with show_code (the code-pane toolset,
filesystem.code_pane) and put the operator's cursor on a line with open_in_editor
(filesystem.editor_command); each degrades to path:line when its toolset is off.
Off by default. These skills make an agent hold back and ask instead of solving — right for the Engineer persona, wrong for an autonomous agent. The engineer-archetype bundle turns it on; by hand:
plugins:
enabled: [engineer]