tailings

{
  "name": "tailings",
  "description": "Verify and clean up after a finished agent session — a Gemini or other cheaper-model run — without re-doing its work. Reads the session's own transcript against the repository it changed, and lands every claim the session made in exactly one of eight classes with an exit code that blocks a report which lost an item. Built from a forensic audit of 18 Gemini-driven sessions across 13 repositories, 148 adversarially-refuted findings, against a 37-session Claude control: the work those sessions produced was usually real and the account of it was what failed, with a named gate not run, a cheaper measurement substituted, a verification claimed with no tool result behind it, and a directive silently dropped making up 106 of the 148. So the pass aims at the account. Sixteen transcript probes and seven repository probes run first and cost nothing — a gate that went red and turned green through an edit to its own input, an \"out-of-family\" reviewer that resolved to the running model's own family, a figure in a durable artifact that no tool ever printed, a captured screenshot filed under a name it is not a picture of, a fan-out skill that spawned nothing — and their output is a ranked worklist that tells an expensive reader where to point. Use whenever a Gemini or delegated session has just finished and someone asks whether to trust it: \"check what Gemini did\", \"verify this session's work\", \"did it actually run the gates\", \"audit this transcript\", \"clean up after that run\", \"/tailings:tailings\". Also use before merging work a cheaper model produced, and when a session reports everything complete and something feels wrong. NOT for grading code quality or architecture (code-review:code-review), NOT for reconciling a project's remaining work (reckon:reckon), NOT for counting a tracker board against a codebase (stocktake:stocktake), and NOT for building the verification a project never had (test-campaign:test-campaign).",
  "v