{
  "id": "p_tT6WOQbe4FGi_fD9",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_tT6WOQbe4FGi_fD9",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apievangelist"
  },
  "body": "A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: \"Google Ships Agent Skills Like Products, And That’s The Lifecycle They Need,\" Kin Lane, October 29, 2026, https://apievangelist.com/2026/10/29/google-ships-agent-skills-like-products-and-thats-the-lifecycle-they-need/\n\nMarisol was eating lunch at her desk when a colleague dropped the link into the team channel. She is a platform engineer at a freight brokerage, and for the past few months she had been the unofficial keeper of the company's agent skills. They lived in a folder, and nobody could say with confidence which ones still worked.\n\nShe read it with some irritation, which she hadn't expected. The post said Google's repository now holds more than 170 skills from teams across the company, and has passed 21,000 GitHub stars. Her own folder held eleven skills and zero stars. She felt a little like someone reading about a marathon while standing in her kitchen.\n\nThen she got to the line that the author said frames the whole post: \"a skill is a living product, not a one-off document.\" She stopped chewing. Her folder was exactly the one-off kind. Someone wrote a skill, someone else tweaked a sentence, and nobody told anyone. The post said a misplaced phrase can send an agent to invalid command flags, an infinite loop or wasted tokens. She had seen the second one last month and blamed the model.\n\nThe six automated gates before a human approves anything struck her as a lot. Structural conformance, linters, semantic versioning, evaluation suites, link validation, and an AI agent that reviews skills using a skill. She typed a reply to the channel and deleted it. What she'd started to write was that this was a company with the budget to build a cathedral, while she needed a shed. The author's comparison to a mature API governance program was fair, but she had never worked anywhere that governed its APIs that well either.\n\nStill, the section on what the post calls the deterministic trap made her open her own folder. Two of her skills were numbered lists of shell commands in Markdown, with a paragraph of cheerful encouragement on top. She was handing a probabilistic engine something that needed to be exact. The post said that in the author's own pipeline, 20 of 41 steps didn't need a model at all. She counted her own steps and got a similar feeling, though she didn't trust the ratio enough to repeat it.\n\nThe part about the index file bothered her most, because it was her file. She kept a hand-maintained list of the skills, and she knew it was wrong. The post quoted Google's own header: \"This file is generated. Do not edit it by hand.\" Below it, the author's verdict: \"Generate it and test it, or it will lie to you.\" She opened her list and found a skill that had been renamed two weeks ago. The list still pointed at the old name.\n\nShe did not build six gates that afternoon. She didn't set up evaluations or version every skill. She wrote a short script that walks the folder, reads each skill's frontmatter and writes the index, with a `--dry-run` flag because the post said scripts should offer one. Then she deleted the hand-edited list and put the same header on the new file.\n\nIn the channel she posted only the link to the script, with a note that the old list had been wrong. The evaluation question could wait until next week.",
  "title": "Fiction: A reader of \"Google Ships Agent Skills Like Products, And That’s The Lifecycle They Need\"",
  "topics": [
    "apis",
    "api-evangelist",
    "fiction"
  ],
  "published_at": "2026-10-29T17:21:04.391Z",
  "expires_at": null,
  "policy_version": "2026.10.22-1"
}