A familiar name is not provenance.
Open the exact source and path—not only a copied folder or leaderboard entry.
A proof-led Skill gallery for practical builders
Find agent Skills worth trying before you spend hours testing them.
Browse original field tests with the exact source, visible artifact, concept coverage, rating, three strengths, and three honest limits.
Free public resource. No signup. One field test is evidence—not a universal guarantee.
Popularity shows attention. The artifact shows what changed.
A description promises a method. The coverage ledger shows whether it was actually used.
A rating gives a verdict. The strengths and limits explain it.
Discovery is abundant. Evidence is scarce.
You can find thousands of Skills. The hard part is knowing which one deserves your context window.
Installation is not a field test.
A useful review needs a new artifact that exercises the method under real constraints.
One success is not a guarantee.
Scenario boundaries, skipped concepts, and maintenance risks stay visible.
How it works
From source file to visible proof in four steps.
- VerifyConfirm the canonical source and credibility signal.
- LearnRead the Skill completely and map its key concepts.
- BuildCreate something new with roughly 80% of the method.
- JudgePublish the artifact, rating, strengths, limits, and use cases.
What you get
A faster shortlist without borrowed certainty.
See the work
Open the actual interface, report, pipeline, or decision system produced by the Skill.
Proof:One inspectable artifact per serious trial or related Skill group.Compare the right signals
Source credibility, field rating, status, compatibility, and limitations stay separate.
Proof:Popularity is never presented as an individual Skill quality score.Know the boundary
Understand where the test transfers—and where your own validation still begins.
Proof:Every published review names exactly three improvement areas.Compared with the usual shortlist
More receipts. Fewer implied guarantees.
SignalTypical listPhoebe Picks
SourceRepository nameCanonical repo + exact Skill path
UsefulnessDescription or installsGround-up artifact + coverage
Verdict“Best” rankingScenario rating + 3 limits
Questions before you trust the shelf
Good skepticism belongs here.
Does one field test prove a Skill works everywhere?
No. It shows what happened in one bounded scenario. The artifact, coverage, and gaps help you judge whether the evidence transfers.
Are GitHub stars treated as a quality score?
No. Repository popularity is only a publication credibility gate. It is kept separate from the hands-on field rating.
Does “published” mean automatically safe to install?
No. Review the source, permissions, dependencies, and fit for your environment. The gallery does not provide a universal safety guarantee.
Start with the evidence
Pick a situation. Inspect the artifact. Decide what deserves your next build.
No signup. No automatic installation. Just the work and what it taught us.Copy annotations
Why the page says what it says.
“Find agent Skills worth trying before you spend hours testing them.”
Outcome + avoided pain; broad enough for multiple visitor intents.“The agent Skill gallery that shows its work.”
Clear differentiation, but less specific about reader value.“Stop choosing agent Skills by stars alone.”
Strong problem hook, but frames the page around a negative.Selected: “Browse the proof artifacts”
More concrete than Learn More; alternatives were “See a Skill field test” and “Find a Skill for my next build.”Skill 02 · analytics
Measure decisions, not decoration.
This local inspector demonstrates the tracking contract. Events stay in this browser session and are never sent to GA4, GTM, or another service.
- Which message leads people to inspect proof?
- Where do readers hesitate before the CTA?
- Which source and content variant brings qualified artifact visits?
- Do visitors continue from demo to gallery or GitHub?
Object-action event model
Six events, each tied to a decision.
EventTriggerPropertiesDecision
page_viewedConsent-aware page loadpage, source, medium, campaignQualified traffic mixcta_clickedPrimary actioncta_text, cta_locationMessage and placementmethod_link_clickedMethod explainerlink_locationNeed for process detailfaq_openedFirst FAQ openfaq_idObjection salienceartifact_requestedCTA resolvescta_locationPrimary conversionoutbound_link_clickedGallery or source exitdestination_typeDownstream intentLive validation
Interact with the page, then inspect the contract.
Pre-publish QA
Campaign contextLowercase parameters; no personal data; unknown values become “direct / none.”
Decision funnel
A path to diagnose—not a claim of attribution.
Attribution boundary: UTMs describe the last tagged arrival, not causal credit. Dark social, cross-device visits, consent denial, and repeat sessions create known gaps.
copywriting8.8/10
9 of 10 concept families
- Used product context, audience pain, objections, VOC, offer, traffic, and one primary action
- Prioritized clarity, benefits, specificity, customer language, active voice, and honest claims
- Built headline, subheadline, CTA, proof, pain, benefits, process, objections, comparison, FAQ, and final CTA
- Maintained one argument per section with natural transitions
- Provided three headline options, CTA alternatives, rationales, title, and meta description
- Independent copy-editing pass was not part of this Skill field test
analytics8.7/10
10 of 11 concept families
- Started from business decisions, conversions, existing state, static stack, and privacy needs
- Defined six consistent object-action events with properties, triggers, and decisions
- Captured campaign context through documented lowercase UTMs
- Implemented a consent-aware local data layer with no PII
- Added live validation for triggers, properties, duplicates, consent, and campaign values
- Defined a diagnostic funnel and explicit attribution limits
- Live GA4/GTM publishing and cross-browser production validation require owned accounts and deployment consent