swarm-advisor
Skill2 starsChanged 19 days agoApache-2.0View on GitHub
When and how to consult your advisor — before committing to an approach, when stuck, before completing risky or irreversible work. Referenced from the `swarm` skill's rule 9a; for orchestrators only — child agents have no advisor.
In this repository26 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/vendor/vercel-react-native-skills/AGENTS.mdAGENTS.md
- swarm-batchingSkill
- swarm-coderSkill
- swarm-debuggerSkill
- swarm-designerSkill
- swarm-mechanicalSkill
- swarm-orchestratorSkill
- swarm-researcherSkill
- swarm-reviewerSkill
- swarmSkill
- swarm-spikeSkill
- swarm-ui-reviewerSkill
- swarm-workflowsSkill
- building-componentsSkill
- expo-design-systemSkill
- expo-native-uiSkill
- graphifySkill
- mobile-android-designSkill
- mobile-ios-designSkill
- ponytail-debtSkill
- ponytail-reviewSkill
- ponytailSkill
- ui-ux-pro-maxSkill
- vercel-react-native-skillsSkill
- web-design-guidelinesSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Only orchestrators have an advisor.
- Very specific: line 10
Follow the swarm skill first.
- Very specific: line 23
If you're asking for sign-off before completed, save your deliverable first: commit, write the file.
- Distinctive: line 26
One decision per consult: Don't fold three unrelated questions into one call.
- Distinctive: line 28
At most one focused follow-up: if the first answer left a real gap.
- Distinctive: line 29
Authoritative evidence wins: When your own evidence and the advice conflict, go with what you can point to a path or line for — but say so …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Consulting your advisor
Only orchestrators have an advisor. If you are a child agent, stop here and escalate to your orchestrator instead.
Follow the swarm skill first; this adds to it.
When to consult
Before committing to an approach when more than one option looks reasonable. When an error keeps recurring or your reasoning has stopped converging. Before a step that's hard to undo — a migration, a delete, a force-push, telling your parent something is done. Orchestrators, debuggers and reviewers consult before every completed. Skip it for routine work that has one obvious way to do it: the advisor is for a real decision, not a rubber stamp.
How to ask
Your advisor doesn't automatically share your context. A native Claude advisor call runs inside your own conversation and sees your full history; every other advisor — and a simulated call for any agent kind — gets a context file the daemon assembles fresh: your brief, recent checkpoints, open requests, and at most a truncated slice of your recent transcript, never the whole thing. Don't rely on either shape. Write your question and focus paths so the file carries everything a reader of only that slice would need:
- What you're deciding, in one sentence.
- The evidence you've gathered, with paths and line numbers — not a paraphrase.
- Your current approach, and the alternative(s) you're weighing.
- The exact question you want answered.
- What happens if the advice turns out wrong — the blast radius.
If you're asking for sign-off before completed, save your deliverable first: commit, write the file. The call takes real time; a durable result survives a dropped session, an unwritten one doesn't.
Decision rules
- One decision per consult. Don't fold three unrelated questions into one call.
- Advice is evidence, not authority. Weigh it against what you've directly verified — a file's actual content beats a plausible-sounding claim about it.
- At most one focused follow-up if the first answer left a real gap. Don't loop.
- Authoritative evidence wins. When your own evidence and the advice conflict, go with what you can point to a path or line for — but say so out loud in your next checkpoint rather than silently picking a side.
- Unresolved, high-impact uncertainty is not yours to guess through. Send it up: your parent, or the user if you have none.
Record the outcome
Your next checkpoint states what you did with the advice: followed it, or declined it and why. This is the only paper trail your parent or a later reviewer has for the decision — don't skip it even when you agreed with the advice.
flowchart LR
A[Agent working its assignment] -- self-contained briefing --> V[Advisor model, read-only, one decision]
V -- one decisive answer --> A
A -- record followed / declined + why --> K[checkpoint]Structure inspired by the advisor skill in scdenney/open-science-skills (CC BY-NC 4.0); this text is original.
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
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.

