← Smart Chess:Train+Learn to winCONTENT HISTORY

Update to Smart Chess:Train+Learn to win

Snapshot Nov 12, 2026 · 12:08 UTC · version 4.0.0

Collection source: downloaded plugin package.

WHAT CHANGED

First saved snapshot

No earlier snapshot is available to establish a change.

Save Smart Chess:Train+Learn to win and check future changes in Following.

Up to 3 plugins free · no card required.

Compare saved observations

Download comparison JSON
Full technical diff · 0 changed fields
Full snapshot data
{
  "description": "Give the user a playable chess puzzle with the solution withheld, optional progressive hints and a conversational review. Use for 'Give me a chess puzzle', mate challenges or requests for another puzzle; do not use for a guided line that should show the moves.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 169
    },
    {
      "relative_path": "references/starter-puzzle.md",
      "size_in_bytes": 1003
    }
  ],
  "name": "chess-puzzle",
  "skill_md_contents": "---\nname: chess-puzzle\ndescription: Give the user a playable chess puzzle with the solution withheld, optional progressive hints and a conversational review. Use for 'Give me a chess puzzle', mate challenges or requests for another puzzle; do not use for a guided line that should show the moves.\n---\n\n# Chess puzzle\n\n1. Use the requested theme, side and level. For an unspecified request, start immediately with a simple mate-in-one; read [the starter puzzle](references/starter-puzzle.md) for a verified default. For later puzzles, choose a different verified position suited to the user's response. Request a FEN if a supplied external reference cannot be verified with available tools.\n2. Establish the objective and verify the position and intended solution before presenting it. Check legal defensive replies before claiming a forced sequence or unique answer. Tool acceptance validates setup legality, not puzzle correctness. If verification is insufficient, simplify the puzzle or describe it as an exploration instead of asserting a forced win.\n3. Call `start-workshop` with `opponent: true`, the user's `side`, legal `position`, a short neutral `title` and concise `instructions` naming the objective and asking the user to press Finished after trying. Use `strict: false` and omit `steps` entirely. Omit solution-bearing notes: `showHint: false` hides arrows but still reveals the step list.\n4. Wait for the user's move, answer, hint request or Finished follow-up. Give only progressively stronger clues when requested, starting with a relevant feature rather than move squares or the solution. Do not reveal an engine principal variation as a subtle hint.\n5. Review the active published moves and outcome, or the user's proposed move explicitly as a proposal. Accept alternative valid solutions. For a failed attempt, explain the concrete missed defence or unmet objective; do not infer failure merely from deviation from your line.\n6. Reveal the solution when requested or when reviewing a concluded attempt, explaining the key idea briefly. Offer another puzzle adapted to the observed attempt, then wait unless another was already requested.\n\nStockfish controls replies; an unscripted multi-move puzzle cannot force a particular defence. Do not use scripted steps to hide a solution or treat a weak engine reply as proof of a forced win. Use available position-labelled analysis honestly; no puzzle-import service, rating, automatic custom-goal grading or permanent results log exists. Avoid repeating the rendered task below the board.\n"
}

SHA-256 of public snapshot: 2fdcd73d1f1f90b7d3e7abb2da1e7d8a1e08be17cb0295fa31e761c2a9e03e3f