product-proposal
Skill18 starsChanged 2 days agoMITView on GitHub
Use when the user needs to choose an undecided Fleet Console feature or UX direction; compare measured evidence and interactive options. Use implementation for approved designs, console-e2e for runtime diagnosis, and design-sweep for recurring design audits.
In this repository13 other files it gives its agents.
- .changelog.d/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- ai-gateway-loop-optimizationSkill
- clean-codeSkill
- console-e2eSkill
- console-handoffSkill
- console-performanceSkill
- design-sweepSkill
- git-worktreeSkill
- learning-harvestSkill
- pr-workflowSkill
- rebase-on-canarySkill
- release-version-updateSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Exercise UX judgment: treat the named solution as a hypothesis.
- Distinctive: line 25
Do not blur the mock with current product behavior.
- Distinctive: line 27
Check token/interaction-grammar changes against applicable doctrine.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Product Proposal
Measure the present, separate the user's named solution from their underlying job, and produce a decision-ready UX proposal. Final product-direction selection and implementation are outside this skill.
Inputs
Preserve one verbatim requirement line, distinguish the user-named solution from the underlying Job-to-be-done, and retain explicit constraints/exclusions. Identify affected surfaces and depth (quick/full); discover uncertain surfaces. quick narrows measurements and uses fewer options; neither depth waives uncertainty labels, interaction verification, or Browser handoff. Never send an already-approved direction back through proposal.
Execution
- Check the handoff path first: before building, confirm this Objective's Fleet Browser can open a task-owned tab and reach the kind of URL you will hand over. A private claude.ai page can show a login screen there, a server bound from this session's shell may be refused, and
file://cannot be opened. Choose a reachable serving path, or record the blocker now and verify the mock with headlessconsole-e2etooling instead. Headless verification and a delivered URL are supporting evidence; they do not complete the Browser handoff in step 5. When you publish the mock as a private Artifact, also keep a local copy with its own<!doctype html>and<meta charset="utf-8">by default, served through the reachable path chosen here, because the Artifact source lacks the skeleton the publisher adds. - Measure the present: invoke
console-e2eon an isolated target build and reproduce the user's actions. That skill and its conditional references own fixtures, boot, and cleanup; do not duplicate path/auth/state-seeding recipes here. Cross-check counts, screenshots, HTTP, and relevantfile:line. Label unmeasured behavior as hypothesis/unknown, not current fact. - Exercise UX judgment: treat the named solution as a hypothesis. Choose only fitting prevention, confirmation, undo/recovery, settings, or arming options. Compare each on the same axes: friction, discoverability, accessibility, implementation cost, product consistency, recovery guarantee. Keep product trade-offs visible.
- Build the mock: invoke
frontend-designusing the target Console's current tokens and applicable design doctrine. Report missing external skills rather than vendoring their workflows. Give each option real interactions, strengths/trade-offs/data risk, a recommendation, and phased delivery. Do not copy model-specific prompts or historical token values. - Verify and hand off: exercise the proposal's interactions in this Objective's Fleet Browser and inspect screenshots. Match mock labels to the target product's actual UI language; explanation follows session language. Open the final proposal URL in a task-owned tab, confirm the intended page loaded, and select that tab for the user. Do not navigate or close pre-existing user tabs. Leave the proposal tab and any owned serving process needed to view it running for the user's direction choice.
- Deliver: measured facts and labeled hypotheses → UX judgment and trade-offs → interactive proposal URL → recommendation. Follow the environment's publication/file and sensitive-data boundaries. A visual proposal is complete only when the verified proposal is open in Fleet Browser for the user to inspect. If Fleet Browser is unavailable, report the blocker and provide the URL without claiming Browser handoff is complete.
Boundaries and completion
- Do not blur the mock with current product behavior. Label it as a proposal and exclude user data/credentials.
- Within the authorized proposal scope, continue reversible isolated mock edits and verification without intermediate approval.
console-e2eowns current-product measurement and cleanup of its verification resources; this skill owns the standalone proposal handoff. Useconsole-handoffwhen delivering an actual Console instance for the user to try. - Check token/interaction-grammar changes against applicable doctrine. Do not silently change product-wide values to solve one panel.
- Once options, evidence, comparison, recommendation, and Browser handoff are complete, stop for the user's direction choice. Do not begin implementation or PR publication.
- For blocked measurement/mock dependencies, report available investigation and incomplete scope. Never fabricate numbers or browser-verification results.
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.

