Kick off seqcli skills install and seqcli mcp run

[nblumhardt-ro]

This PR starts a new feature branch for integrated search and query tools, and a companion agent skill. We're exploring this approach for a few reasons -

• seqcli already provides a stack of useful infrastructure that this and other MCP tools can build on,
• is already bundled with the Seq installer on Windows,
• is updated in step with Seq releases by the Datalust team, and
• can benefit from the tooling we have for generating documentation, schemas, syntax references and so on.

The code currently in this PR pulls together the main ingredients. In particular, it leans very heavily on the Seq query language, and front-loads a dense syntax and function reference via the skill. Our intention is that sessions can smoothly ramp up complexity in their interactions with Seq, without needing to switch techniques once they're outside of the pre-baked skill parameters. The skills also jump through some hoops to keep all representations firmly within Seq's query and literal syntax, to mitigate some common pitfalls where agents gravitate to JSON or SQL representations.

From here there's still a lot to do, but the results seem promising so far.

If you're keen to pull the branch and try it out at any point, to get the skill set up, try:

seqcli skills install -a <your agent> --global

The MCP server is:

SEQCLI_CONNECTION_SERVERURL=<your server URL>
SEQCLI_CONNECTION_APIKEY=<your API key>
seqcli mcp run

(For the latter, dotnet build it and configure your agent to invoke dotnet with SeqCli.dll - the dotnet run command produces output that will trip up the STDIO MCP protocol.)

The EBNF grammar in SKILL.md was initially produced from our parser by Claude, which was also used for table formatting; the ugly C# is all mostly mine. Edit I'm filling in some test infra gaps and setting up an end-to-end MCP test case using Claude. Aim for the final PR is to write by hand anything that lays down design decisions/foundations, and fill out boilerplate/broader tests using assistance.

[nblumhardt]

Some .gitignore entries needed; I'll clean this all up a bit :-)

[nblumhardt-ro]

I think this is as far as I should go before we sync/merge/plan.

The major unanswered question is how signals should be handled; I don't think agents will use signals unless they're extremely strongly motivated to, and I also expect a lot of footguns there, so it's a decent chunk of work to continue with.

As far as Seq interactions and the query language go, though, it's great - now seeing close to zero search/query usage mistakes in my (very informal) evals. We'll still definitely need to feed back through a lot more skill tweaks based on real-world usage.

A future version should probably integrate the schema tool with search results, so that search results always print all newly-observed property accessors before or after the results themselves. Claude is extremely unreliable when it comes to calling the schema tool, except when the prompt suggests planning before executing.

Right now the intended setup flow is:

dotnet tool install --global seqcli
seqcli config -k connection.serverUrl -v https://seq.example.com
seqcli config -k connection.apiKey -v 1234567890abc
seqcli mcp install -a claude --global
seqcli skill install -a claude --global

Desperately needs a squash 😊

[KodrAus]

Just a few nitpicky comments here and there, but let's keep this rolling along 👍