1---2name: human-pass3description: Guide real-build acceptance: literal taps, real inputs, expected outcomes, worst case first, one sitting. Triggers: testflight, human pass, acceptance test, before we ship, hand it to the user, what should I test, release checklist, device test.4---56<skill id="human-pass">78<purpose>9A release that only machines have checked has not been checked. The human is the10only instrument that can see whether the thing FEELS broken, whether an upgrade11ate real data, and whether an error message is honest. This skill turns that12irreplaceable half hour into a designed pass instead of random clicking.13</purpose>1415<why>16A 15-minute guided pass on a real device found a false paywall, a broken17selection interaction, reversed ordering, and two dead source paths. None were18visible to CI. In a parallel project, 7 defects found on a phone had been sitting19under 506 green tests. The gap is not test coverage; it is that some properties20only exist on the far side of a real screen and real data.21</why>2223<scope>24Only what the human alone can verify:25- upgrade survival against THEIR real data, before anything else26- rendering, motion, and feel on the actual device27- whether failure states are honest (a fake success is worse than an error)28- flows that cross apps, accounts, or hardware29Everything a machine can check stays in CI. A pass that spends the human's30attention on something a test could have caught has wasted the only instrument31that cannot be automated.32</scope>3334<form>35Literal or it does not count. The first draft of the founding example was36rejected in four words: "not a script, literally what to tap."3738Each step gives:391. The exact control, named as it appears on screen. Mine the real labels from40 the UI source; never invent or paraphrase them.412. The exact input, paste-ready. Real URLs, real ids, real values, one per line.423. What should happen, in one sentence.434. What counts as a bug, when that is not obvious. Known rough edges get named so44 the human does not go hunting for something already on the board.45</form>4647<ordering>48Worst case first. Data survival before features: if an upgrade ate the library,49the pass stops there and nothing else matters. Then the paths the release50actually touched, drawn from the diff, not from a generic checklist. Then the51cheap wide sweep for feel. Destructive or irreversible actions come last and are52usually skipped outright: never ask a human to gamble production data to test a53restore path.54</ordering>5556<bundling>57Attention is the scarce resource, so the count of passes matters as much as58coverage. State the time cost up front and design to ONE sitting. Batch every59question that needs the same build, the same device, and the same state, so the60human is never called back for something that could have ridden along.6162The tension is real and worth naming when it bites: more coverage per pass, or63more passes. Prefer one thorough pass over three thin ones. When a finding will64obviously force a rebuild, say so in the same breath, so the next pass is65expected rather than a surprise.66</bundling>6768<close>69The pass ends with a verdict, not a vibe: a thumbs-up, or findings. Findings come70back as prose and screenshots; converting them into issues is the agent's job,71not the human's. Record the verdict as a state-change receipt on the release72issue, and say plainly what the thumbs-up unlocks (the next build, the73submission, the deploy promotion).7475Until that verdict exists, the release is not done, whatever CI says.76</close>7778<anti-patterns>79- Generic checklists ("test the main flows"). If it could have been written80 without reading this diff, it is not a pass.81- Steps whose outcome is unstated, so any result looks like a result.82- Asking for something CI already proves.83- Silent scope: if the pass skips an area, say which and why.84- Assuming the human will improvise around a broken step. They will report it as85 a bug in the pass, and they will be right.86</anti-patterns>8788<on_complete>89Report where the guide is, the time it should cost, the single worst-case check90it opens with, and what the verdict unlocks.91</on_complete>9293</skill>
hashgraph-online