Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 10 additions & 5 deletions build-mcpb/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -402,19 +402,24 @@ Extract the server's tool surface:

### 6b: Draft SKILL.md

Generate a **draft** embedded skill file:
Generate a **draft** embedded skill file - or edit the generic scaffolded one, if it exists:
- **Python:** `src/mcp_<name>/SKILL.md`
- **TypeScript:** `src/SKILL.md`

No frontmatter — this is pure Markdown. The draft should contain:
The draft should contain:

1. **Tool selection table** — each tool with a one-line description and when to use it
2. **Context reuse rules** — which tool outputs feed into subsequent calls
3. **Multi-step workflow patterns** — 2–3 composed workflows showing how to chain tools

The frontmatter should contain the name and description of the skill.

Example structure:
```markdown
# <Service> MCP Server — Skill Guide
---
name: mcp-<name>-service
description: Provides knowledge of how to use MCP <name> most effectivelly. It's loaded into the agent's context when running the MCP.
---

## Tools

Expand Down Expand Up @@ -443,7 +448,7 @@ Example structure:

### 6c: Review with Contributor

The embedded skill created so far is inherently opinionated — it shapes how an LLM uses the server. The contributor understands the target API's nuances better than this pipeline can: which tool combinations are most valuable, what context flows are non-obvious, and what real workflows users will care about.
The embedded skill created so far is inherently unopinionated — it was created generically based on common knowledge and basic patterns. The contributor might have a specific idea of how t use the MCP already in mind though. So it's important to make this Step 6 interactive and request confirmation before moving on to the next Step.

**Show the draft to the contributor and ask for their input.** Present the full SKILL.md content and ask:

Expand All @@ -458,7 +463,7 @@ The embedded skill created so far is inherently opinionated — it shapes how an
- Any context reuse rules that aren't obvious from the tool signatures?
- Should any workflows be removed or reframed?

=> Let me know what to change, or approve to continue.
=> Let me know if it needs change, or approve to continue.
```

Iterate with the contributor until they approve the SKILL.md. This may involve multiple rounds — the contributor might add domain-specific workflows, adjust tool selection guidance, or refine context reuse rules that only someone familiar with the API would know.
Expand Down
2 changes: 1 addition & 1 deletion nimblebrain-contributor/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -210,7 +210,7 @@ Present the results clearly — summarize what's open, what's recently closed, a
After presenting, offer a natural next step:
"Want to pick up something new, or jump back into what you were building?"

If new work → Path A. If continuing a build → tell them to type /build-mcpb to pick up where they left off.
If new work → Path A. If continuing a build → invoke `/build-mcpb` in this same session to pick up where they left off.

### Path D: Exploring the Ecosystem

Expand Down