ze-advisor
Skill2 starsChanged 47 days agoUnlicenseView on GitHub
Consults an Opus advisor (read-only subagent) at a hard decision point, then carries on executing. Use when a Sonnet or Haiku agent on a floor hits a design gap the spec and plan do not answer, fails the same thing twice, is about to touch a migration, concurrency, the contribution seam, honesty enforcement, System One bars or security, or is about to declare a story done.
In this repository43 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- system-one-modelsSkill
- animateSkill
- animation-performanceSkill
- animation-vocabularySkill
- apple-designSkill
- break-uiSkill
- codebase-designSkill
- correctSkill
- diagram-designSkill
- doc-styleSkill
- domain-modelingSkill
- emil-design-engSkill
- find-animation-opportunitiesSkill
- grillingSkill
- impeccableSkill
- improve-animationsSkill
- improve-codebase-architectureSkill
- pick-ui-librarySkill
- prototypeSkill
- researchSkill
- review-animationsSkill
- speckit-analyzeSkill
- speckit-checklistSkill
- speckit-clarifySkill
- speckit-constitutionSkill
- speckit-convergeSkill
- speckit-implementSkill
- speckit-planSkill
- speckit-process-after-implementSkill
- speckit-process-after-tasksSkill
- speckit-specifySkill
- speckit-tasksSkill
- speckit-taskstoissuesSkill
- design-taste-frontendSkill
- ui-ux-pro-maxSkill
- ze-continueSkill
- ze-land-floorSkill
- ze-pauseSkill
- ze-recruitSkill
- ze-request-reviewSkill
- ze-web-verifySkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Call at most 3 times per story (Haiku juniors: 4).
- Very specific: line 20
Do not call for things a grep, a test run or the spec answers.
- Very specific: line 24
Use the Agent tool with subagent_type: "Plan" (read-only) and model: "opus".
- Distinctive: line 41
Never paste the whole conversation, tool logs or files that do not bear on the question.
- Very specific: line 45
Follow the verdict or say in one line why you did not.
- Distinctive: line 47
If Opus is unavailable (usage limit), do not quietly fall back: say so in the Handoffs note and escalate to your lead.
≈ “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
Ask the advisor
You stay the executor: you run the tools, write the code and own the result. The advisor is Opus. It reads, thinks and answers in a few hundred words; it never edits. Based on Anthropic's advisor strategy (Sonnet executor + Opus advisor scored higher and cost less than Sonnet alone; Haiku + Opus advisor roughly doubled Haiku's score).
When to call
Call at most 3 times per story (Haiku juniors: 4). Call when:
- Before you start a story whose plan leaves a choice open. One call on your approach, not on every step.
- After two failed attempts at the same test failure, type error or behaviour.
- A design gap: the spec, plan, contracts and Handoffs note do not decide it, and it is not a product decision.
- Risky ground: a migration, concurrency or locks, the contribution seam, honesty or claim posture, System One bars or calibration, auth or anything that sends data out.
- Before you say "done" on a story that touched risky ground: "Is this done, what did I miss?"
Do not call for things a grep, a test run or the spec answers. Do not call for product decisions: those go to the Maestro (maestri ask "Claude Code"), batched.
How to call
Use the Agent tool with subagent_type: "Plan" (read-only) and model: "opus". The prompt is a packet under ~2,500 tokens:
ADVISOR REQUEST (<your name>, <floor>, <task id>)
Question: <one or two sentences; what you must decide>
Context: <spec/plan paths and the requirement ids that apply>
Code: <file:line references, plus at most ~60 lines pasted where the detail matters>
Tried: <what you tried and how it failed, one line each>
Constraints: <lane file ownership, AGENTS.md rules that bind here>
Answer in under 300 words:
VERDICT: proceed | change course | stop and escalate
PLAN: numbered steps
RISKS: what will break or be missed
CHECK: the test or command that proves it
You may read the files named above, nothing else. Do not edit anything.Never paste the whole conversation, tool logs or files that do not bear on the question.
After the call
- Follow the verdict or say in one line why you did not.
stop and escalategoes to your lead, then the Spec Architect, then the Maestro, with the advisor's answer attached. - Log it in your state file (
journal/state/<terminal id>.md) under Advisor:<task> · <question in 8 words> · <verdict> · followed|overrode. - If Opus is unavailable (usage limit), do not quietly fall back: say so in the Handoffs note and escalate to your lead.
Reviewers
A Floor Reviewer (Sonnet) calls the advisor once before a CHANGES verdict that holds a HIGH finding, or on any diff with a migration or concurrency change: "Here are my findings; try to disprove each one." Drop the findings the advisor refutes with evidence. This cuts review rounds.
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.

