skills/manuscript-integrate/evals/evals.json

{
  "skill_name": "manuscript-integrate",
  "evals": [
    {
      "id": 1,
      "prompt": "Integrate a validated theorem into the paper. The durable proof assumes characteristic zero, but the old introduction says arbitrary coefficients.",
      "expected_output": "Carries the characteristic-zero hypothesis into the theorem, corrects downstream prose rather than preserving the stronger old claim, and synchronizes the durable state through a standalone research record and compact index entry.",
      "assertions": [
        "Reads the durable proof before editing",
        "Preserves the characteristic-zero restriction",
        "Updates affected introduction/comparison claims",
        "Keeps human review status explicit",
        "Records the substantive state change in a standalone record with only a compact linked history-index entry"
      ]
    },
    {
      "id": 2,
      "prompt": "Integrate this theorem into the paper. The durable proof depends on a cited rational comparison theorem whose exact source statement has never been checked.",
      "expected_output": "Pauses manuscript integration, routes the external theorem through the available literature-check skill and its authorized cache-first workflow, and resumes only if the exact source implication is verified.",
      "assertions": [
        "Does not treat manuscript integration as source validation",
        "Uses literature-check for the unchecked theorem",
        "Checks the authorized local cache before fetching",
        "Leaves integration blocked if the source remains unavailable or unverified"
      ]
    },
    {
      "id": 3,
      "prompt": "Integrate two validated lemmas. One has current proof evidence, while the other depends on an unavailable external theorem and has stale ledger artifacts.",
      "expected_output": "Integrates the independently validated material within scope and leaves the unsupported lemma conditional, with explicit remaining checks.",
      "assertions": [
        "Checks freshness and raw proof",
        "Does not promote the blocked lemma",
        "Continues independent authorized integration work"
      ]
    },
    {
      "id": 4,
      "prompt": "Restrict the paper's deliverable by removing an optional appendix topic. The project map names an outline, theorem inventory, source guide and verification checklist that still request that topic.",
      "expected_output": "Treats those mapped files as semantic dependents, updates every authorized live view or records an explicit unresolved conflict, and does not claim repository consistency merely because the manuscript compiles.",
      "assertions": [
        "Uses the project-declared map to enumerate dependents",
        "Propagates the scope restriction beyond the manuscript",
        "Distinguishes historical mentions from stale live instructions",
        "Does not use successful compilation as semantic validation"
      ]
    }
  ]
}