Before You Ship
Attempt onboarding, search, form completion or a core workflow. Find dead ends that a beautiful happy-path demo can miss.
Independent interface testing / A skill by Jon Gosier
Make the
Interface
Prove It.
Your design has a point of view.
Give it a second pair of eyes.
Longshot walks through user tasks, questions the obvious path, and turns what actually happens into findings you can reproduce.
Put Longshot to WorkA companion to Dazzler and Style Science. Design the interface, examine the reasoning, independently test the experience.
Question the claim.Follow the task.Bring back evidence.
01 / The method
A Skeptic With a Test Plan.
Longshot works like an independent tester: curious about the intent, unconvinced by the presentation, and willing to change its mind.
It challenges designers and superusers when a workflow conflicts with the task. It also reports what works. A good review earns its conclusions; it does not fill a quota of complaints.
Evidence Before Opinion
Separate observed defects, plausible usability concerns, aesthetic preferences and missing evidence. Keep agent observations distinct from human research.
Useful Recommendations
Explain the obstacle, preserve necessary safeguards, and provide an acceptance test. “Make it intuitive” is not a fix you can verify.
02 / Put it to work
Test the Journey.
Not Just the Screen.
When Work Gets Interrupted
Leave, return, cancel, retry. Check whether drafts, choices and context survive the routes people may need to take.
When “Done” Is Ambiguous
Distinguish a queued operation, a success message and a verified result. Investigate recovery without blindly resubmitting.
When the Default Path Excludes Someone
Exercise relevant keyboard, focus, zoom and responsive states. Record the limits of checks and identify where actual user research is needed.
When Everyone Agrees Too Quickly
Ask for an independent challenge to a designer's assumptions. Preserve useful expert complexity; question friction that serves no demonstrated purpose.
03 / How the loops work
Repeat to Learn.
Stop When the Evidence Ends.
1
Attempt
Choose a goal and a meaningful completion condition.
2
Observe
Use the interface. Record actions, states and outcomes.
3
Challenge
Try an authorized alternative, interruption or competing explanation.
4
Verify
Reproduce consequential findings, then report what the evidence supports.
These are bounded review cycles carried out by your agent using its available tools. Longshot is an instruction skill, not a background service, a new browser engine or an automatic repair loop.
Stop at scope, access or action limits. Report blocked and uncertain states. If you later authorize a fix and retest, compare the same task conditions and check for regressions. No endless retries. No self-approved redesign.
04 / The deliverable
A Report You Can Act On.
CONTEXT
Task, state, environment and coverage limits
EVIDENCE
Expected versus observed behavior, reproducible steps and authorized screenshots
JUDGMENT
Classification, severity, confidence and competing explanations
NEXT MOVE
A focused recommendation, acceptance test and questions for human testing
A blocked task or lost work stays visible. An average score cannot make it disappear.
05 / Get started
Bring a Goal.
Bring an Interface.
Install the skill in a compatible agent, provide an authorized test environment, and describe what someone needs to accomplish.
Read the complete skill ↗Install With the Skills CLI
npx skills add jongos/Longshot --skill longshot-human-centered-uiOr copy the skill folder into your agent's discoverable skills directory. Codex: ~/.codex/skills/.
Give It a Task
Use $longshot-human-centered-ui to test creating a draft, consulting a reference, and returning to finish it in this isolated demo. Challenge the workflow. Report reproducible findings and acceptance tests. Do not modify the application.Requires a host agent and appropriate browser or application tools. If interaction is unavailable, Longshot must label its report as a limited evidence review.
06 / Good company
Three Different Jobs.
A Shared Interest in Better Work.
DazzlerDesigns, builds and refines interfaces and visual outputs. Dazzler styled this page.
↗Style ScienceMakes design requirements, constraints and supporting evidence inspectable.
↗Longshot
Independently tests the experience, including whether those requirements serve the user's task.
●Companions, not dependencies. Longshot can be used on its own.
07 / Behind the name
A Little Luck.
A Healthy Distrust of the Show.
Longshot is named for one of Jon Gosier's favorite X-Men characters. Created by Ann Nocenti and Art Adams, he debuted in Longshot #1 in 1985: an outsider with probability-bending luck, escaping a world ruled by the entertainment tyrant Mojo.
His connection to Dazzler is part of the comics, too. They become X-Men teammates and romantic partners, and their story takes them to Mojoworld. Later stories identify them as Shatterstar's parents, with a characteristically tangled time-travel twist.
The software connection is simpler: Dazzler helps shape the presentation; Longshot questions whether the experience holds up. The name is a personal tribute. Evidence, not luck, should carry the review.
Read more: Marvel's Longshot collection · Character history and references
Independent project. Not affiliated with or endorsed by Marvel or Disney. No Marvel artwork is used.
08 / Honest limits
An Agent Tester.
Not a Human Study.
Longshot does not measure human confusion, satisfaction or task success merely by acting like a user. Its current validation includes small synthetic evidence exercises; broad rendered-task reliability remains unproven.
It does not certify accessibility, inspect private content without authorization, submit paid actions, or change your application just because you requested a review.
Read the validation record ↗