adityamparikh opened a new pull request, #281:
URL: https://github.com/apache/solr-mcp/pull/281

   Closes #279
   
   ## What
   
   Adds `src/main/resources/skills/solr/`, an [Agent 
Skill](https://agentskills.io/specification) about this server's tools:
   
   - `SKILL.md`: which tool does what, and what to know before writing data 
(collections created with defaults share the `_default` configset, document ids 
and re-sending, how field names and values are stored, commits).
   - `references/indexing.md`: what each `index-*` tool does to the data, how 
responses read, partial failures.
   - `references/troubleshooting.md`: the server's error messages, what each 
means, the next call.
   - `references/solr-versions.md`: what differs across Solr 9.4 to 10 as seen 
through the tools.
   
   README, `docs/FAQ.md` and `AGENTS.md` describe the skill. The FAQ's "thin 
skill for query and faceting know-how" now reads as server-specific guidance.
   
   ## Why
   
   The tools give an agent safe operations but not the behaviour behind them: 
ids that decide whether a re-send duplicates, a schema shared across 
collections, a response that does not say which documents failed. The skill 
holds that. It does not teach Solr itself; query syntax, analyzers and schema 
design are Solr Reference Guide material, which an LLM can reach through its 
own knowledge, web search or a docs-lookup tool.
   
   It also avoids restating the tool descriptions, prompts and server 
instructions, so the prompts are unchanged.
   
   ## Tests
   
   `SolrSkillFormatTest` parses the frontmatter and checks the Agent Skills 
rules (allowed keys, `name` equals the directory, description at most 1024 
characters, `metadata` is string to string), `SKILL.md` under 500 lines and 
about 5000 tokens, every reference named in `SKILL.md`, every relative link 
resolving, no `scripts/`, and the SEP-2640 limits of 512 files and 16 MiB.
   
   The claims about ids, generated ids on re-send and array-valued guessed 
fields were checked against a Solr 9.9 container. `./gradlew build` passes: 471 
tests, 0 failures, 0 skipped.
   
   Serving the skill to clients over MCP is #280.
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to