Add schema validation when loading .thinktank result files to prevent cryptic crashes

[that-github-user]

Problem

apply.ts and list.ts load .thinktank/latest.json with no schema validation:

const result = JSON.parse(fs.readFileSync(..., "utf-8")) as EnsembleResult;

If the file is truncated (e.g. disk full mid-write), written by an older version, or manually edited, JSON.parse throws a generic syntax error and the user sees:

Error: No results found. Run thinktank first.

...with no indication of what actually went wrong.

Solution

Add lightweight schema validation using the existing TypeScript types. A minimal runtime check (not necessarily Zod) would verify:

• result.agents is a non-empty array
• Each agent has id, status, diff
• result.recommendation is a valid agent id

function assertValidResult(data: unknown): asserts data is EnsembleResult {
  if (!data || typeof data !== "object") throw new Error("Invalid result file: not an object");
  const r = data as Record<string, unknown>;
  if (!Array.isArray(r.agents) || r.agents.length === 0) {
    throw new Error("Invalid result file: missing or empty agents array");
  }
  // ... etc
}

Acceptance criteria

• Loading a truncated or empty JSON file shows an actionable error message (e.g. "Result file appears corrupted. Re-run thinktank to regenerate.")
• Loading a result from a different schema version shows a version mismatch warning
• apply, list, compare, and stats all use the same validation path
• Unit tests cover: truncated JSON, empty file, missing agents field, missing recommendation

[that-github-user]

Fixed in #126. parseAndValidateResult() validates required fields. All commands skip invalid files with warnings. 24 new tests. Generated by thinktank Opus (+3 Copeland).