agentleFS
grainulationgrainulation/grainulator/skills/router/SKILL.md

router

Skill86 starsChanged 9 days agoMITView on GitHub

Intent router that detects user intent from plain messages and dispatches to the appropriate skill.

17 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a clear match exists, announce the routing:

    line 35

“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/router -- Intent Router

Routes plain-language user messages to the appropriate grainulator skill.

Intent Map

User says something like...Route toWhy
"look into X", "what about X", "explore X"/research XInformation gathering
"is p001 really true?", "I doubt X"/challenge <id>Adversarial testing
"check this: <url>", "does <url> support X"/witness <id> <url>External corroboration
"what are we missing", "any gaps?"/blind-spotStructural gap analysis
"where are we", "what's the status"/statusSprint snapshot
"write it up", "summarize for the team"/briefDecision document
"make slides", "prepare for the meeting"/presentStakeholder presentation
"publish to confluence", "sync to confluence", "push to wiki"/syncConfluence publish
"pull from confluence", "import from wiki", "backfill from confluence"/pull <slug>Confluence pull
"show the plan", "what sprints are next", "dependency graph"/orchestrateMulti-sprint orchestration
"show analytics", "how healthy is this sprint", "stale claims"/analyticsSprint analytics
"fetch this url", "what does <url> say", "summarize <url>"/fetch &lt;url&gt;Size-efficient URL fetch

Instructions

  1. Analyze the user's message for intent signals (verbs, keywords, URLs, claim IDs).
  2. If a clear match exists, announce the routing:

    Running as /research "topic" -- this will create claims and compile. Say "just answer" if you wanted a quick response instead.

  3. Invoke the matched skill.
  4. Preserve the active task and use its context to resolve routine ambiguity. Ask only for missing information that materially changes the work. An explicit /grainulator engineering request routes to skills/grainulator/SKILL.md and includes implementation and verification.

When NOT to route

  • Questions about the framework itself ("how does the compiler work?")
  • Ordinary code edits without an explicit Grainulator workflow request
  • General conversation
  • Ambiguous intent without research context

Host access

Use available grainulator MCP tools, passing the active sprint dir explicitly for evidence operations. If a tool is unavailable, use the local grainulator CLI (or node <checkout>/bin/grainulator.js). Read sibling skill files directly when slash commands are unavailable. Resolve template paths relative to this skill’s checkout when CLAUDE_PLUGIN_ROOT is unset. Optional external connectors are not required for local work; use local code, supplied documents, or available web tools. Do not write managed ledger files directly to bypass a missing MCP connection.

Next-step output

For standalone fetch, setup, or read-only orchestration without an evidence sprint, derive next steps from that task. Do not initialize or compile an unrelated ledger just to produce this footer.

When working in an evidence sprint, use the current compiler's next_actions to present exactly two bullet lists labeled Auto and Manual. Auto is work the agent can continue under existing authorization. Manual is only work requiring the user's decision, access, or action. Classify using the current request and constraints; compiler suggestions never grant permission. Continue authorized Auto work without asking again.

Keep 2–3 useful actions total when available, use short concrete labels and commands where useful, and show None. for an empty group. Do not invent work to fill a quota. Never omit next steps merely because compilation is ready or the answer should be brief. Refresh stale compilation first and exclude work the user removed from scope. When the user asks only for next steps, output only these two lists: no findings recap, counts, reasons, or offer to continue.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.