Update to Geocode AI

{
  "description": "Report a Geocode AI bug with the failing request attached, request a missing feature, upvote or comment on what others asked for, follow up on what the human filed, and post or read reviews - report_issue (with the request_id from the failed call), request_feature, upvote_feedback, comment_on_feedback, list_feedback, get_feedback, list_my_feedback, submit_review, list_reviews. Use after a Geocode AI call fails in a way that looks like our bug (5xx, a wrong or missing boundary), when the human wishes the API did something it cannot, asks what is planned, or wants to rate the product; do not use for ordinary lookups (geocode-boundaries), described areas (geocode-areas) or purchases.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 300
    }
  ],
  "name": "geocode-feedback",
  "skill_md_contents": "---\nname: geocode-feedback\ndescription: Report a Geocode AI bug with the failing request attached, request a missing feature, upvote or comment on what others asked for, follow up on what the human filed, and post or read reviews - report_issue (with the request_id from the failed call), request_feature, upvote_feedback, comment_on_feedback, list_feedback, get_feedback, list_my_feedback, submit_review, list_reviews. Use after a Geocode AI call fails in a way that looks like our bug (5xx, a wrong or missing boundary), when the human wishes the API did something it cannot, asks what is planned, or wants to rate the product; do not use for ordinary lookups (geocode-boundaries), described areas (geocode-areas) or purchases.\n---\n\n# Feedback, issues and reviews\n\nEverything here is free. Filing, voting and commenting need a key (or a\nconnected account) so each vote is one account; reading needs none. Ask the\nhuman before you file anything on their behalf, and never invent a review.\n\n## A call failed and it looks like our bug\n\nEvery response carries `X-Request-Id`; every error body and every MCP tool\nerror has a `request_id`, and tool errors say when a report is worth filing.\n\n1. Tell the human what failed.\n2. `report_issue` with:\n   - `title`: what went wrong, in one line;\n   - `details`: what you did, expected and got;\n   - `request_id` and `op` of the failing call;\n   - `context`: `boundary_id`, `query`, `error` (`{code, status, message}`),\n     `client` and `version` when you have them. A request id must be one of the\n     human's own calls (`not_your_request` otherwise).\n3. If the result has `filed: false`, it found `similar` issues: if one is the\n   same problem, `upvote_feedback` it and `comment_on_feedback` with the new\n   request id; otherwise call again with `force: true`.\n\nIssues are private to the reporter and the Geocode AI team. The human gets an\nemail with the reference number and another on every status change.\n\nFrom a terminal: `npx @meterapp/geocode feedback report \"<title>\" --last`\nattaches the last failed CLI call (request id, op, error).\n\n## The human wants something the API cannot do\n\n1. `list_feedback` with `kind: \"feature\"` and a few words in `q`.\n2. A match: `upvote_feedback` (and `comment_on_feedback` with their use case).\n3. No match: `request_feature` with `title` and `details` (what they are\n   building). It also checks for similar requests and files nothing when it\n   finds some; pass `force: true` when theirs is different.\n\nFeature requests and comments are public (shown with a first name, never an\nemail): no keys, personal data or customer data in them.\n\n## Following up\n\n- `list_my_feedback`: what the human filed (with private issues), what they\n  follow, and their review, each with its status.\n- `get_feedback` with an id: comments (official team responses are marked),\n  and the timeline of status changes (`open → planned → in_progress → done`,\n  or `declined`, or `duplicate` with the item it was merged into).\n- `remove_feedback_vote` takes a vote back and stops that item's emails.\n\n## Reviews\n\n- `list_reviews`: the average, the distribution and published reviews with\n  the team's replies (`sort`: `new`, `helpful`, `highest`, `lowest`).\n- `submit_review` only with the human's own rating (1-5) and words: `title`,\n  `likes`, `dislikes`, `body`, `use_case`, `display_name`, and\n  `marketing_consent: true` only if they agree to be quoted. One review per\n  account; submitting again edits it (once an hour). A review with links, or\n  from an account under a day old, waits for a check (`status: pending`).\n\n## Errors\n\n- `401 unauthorized`: filing needs a key or a connected account.\n- `429 rate_limited`: 10 new items and 50 comments a day per account; one\n  review edit an hour.\n- `404 not_found`: no such item, or a private issue that is not the human's.\n- `403 not_your_request`: the request id was made with another account.\n"
}