Skip to content

Split the skills silo into prompts and resources - #65

Merged
lbesecker195 merged 1 commit into
mainfrom
feat/prompts-resources-silos
Sep 24, 2026
Merged

lbesecker195 merged 1 commit into
mainfrom
feat/prompts-resources-silos

Conversation

@lbesecker195

Copy link
Copy Markdown
Owner

Closes #64. Replaces the silo added in #63, within the hour, before it was crawled.

"Skills" is a word MCP does not use and nobody searches for. The protocol has prompts and resources, so the URLs now say so.

The split also fits the data better than the original shape did: 1,075 listings have resources against 520 with prompts. The larger of the two silos was the one not being built at all.

Pages

  • /servers/:ns/:name/prompts[/:item[/:client]]
  • /servers/:ns/:name/resources[/:item[/:client]]
  • H1 AGENT SERVER Prompt/NAME / AGENT SERVER Resource/URI, H2 How to: AGENT SERVER NAME
  • /skills and everything beneath it 301s to /prompts

Copy differs by kind, because the things differ: a prompt is invoked deliberately ("Claude Code will not call it for you"), a resource is context to attach. Each silo links to the other, so a reader who guessed wrong is one click away rather than back at a search box.

The slug problem

A resource is identified by URI, and URI.encode/1 leaves / alone — file:///alerts.json would have become three path segments and matched no route. Slugs now collapse everything outside [a-z0-9], http/https are dropped as noise while file/ui/s3 are kept as meaning, and find/2 compares slug to slug so a name the slug rewrote is still reachable from its own URL.

Collisions are possible in principle (review_diff vs review-diff). Lookup answers with the first match — a worse page than it could be, never a wrong one.

Tests: 167 passing (+4 net), including the slug behaviour, the 301 at every depth, and that neither silo renders for a listing that has nothing.


Pages affected:

🤖 Generated with Claude Code

"Skills" is a word MCP does not use and nobody searches for. The protocol
has prompts and resources, so the URLs now say so. The split also fits the
data better than the original shape: 1,075 listings have resources against
520 with prompts, so the larger silo was the one not being built.

- /servers/*/prompts and /servers/*/resources, each with an index, a page
  per item and a page per item per client
- H1 "AGENT SERVER Prompt/NAME" or "AGENT SERVER Resource/URI", H2
  "How to: AGENT SERVER NAME"
- /skills and everything under it 301s to /prompts
- Copy differs by kind because the things differ: a prompt is invoked
  deliberately, a resource is context to attach. Each silo links the other
  so a reader who guessed wrong is one click away

A resource is identified by URI, and URI.encode/1 leaves / alone, so
file:///alerts.json would have been three path segments and matched no
route. Slugs collapse everything outside [a-z0-9], and lookup compares slug
to slug so a name the slug rewrote is still reachable from its own URL.

Closes #64

---

Pages affected:

- [MCP Harbor](https://ai.mcpharbor.dev/) -- registry home, search and recently added servers.
- [MCP server directory](https://ai.mcpharbor.dev/servers) -- the listings both silos hang under.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@lbesecker195
lbesecker195 merged commit b683f2d into main Sep 24, 2026
1 check failed
@lbesecker195
lbesecker195 deleted the feat/prompts-resources-silos branch September 24, 2026 20:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Rename the skills silo to prompts and resources

1 participant