Update to Mathbox

{
  "description": "Reconcile a mathematical research repository's current claims, proofs, computations, status, literature dependencies, and failed routes, then recommend the next bounded research moves. Use only when the user asks for a project review, weekly/monthly retrospective, prioritization, a prose project handoff, or “what should I do next?”. Do not use to write a program closeout, migrate research history, operate a .mathbox ledger, or generate its dependency-aware handoff. Default to read-only.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 305
    },
    {
      "relative_path": "evals/evals.json",
      "size_in_bytes": 5781
    },
    {
      "relative_path": "evals/trigger-evals.json",
      "size_in_bytes": 1001
    }
  ],
  "name": "research-retrospective",
  "skill_md_contents": "---\nname: research-retrospective\ndescription: >-\n  Reconcile a mathematical research repository's current claims, proofs, computations, status, literature dependencies, and failed routes, then recommend the next bounded research moves. Use only when the user asks for a project review, weekly/monthly retrospective, prioritization, a prose project handoff, or “what should I do next?”. Do not use to write a program closeout, migrate research history, operate a .mathbox ledger, or generate its dependency-aware handoff. Default to read-only.\n---\n\n# Research retrospective\n\nProduce a decision-quality view of the project, not a chronological summary.\nDefault to no edits unless the user asks to reconcile files.\n\n## Establish authority\n\n1. Determine repository root and applicable instructions.\n2. Resolve charter, live status, claims, conventions, literature ledger, durable\n   proofs, computations, verification, research-history index, and detailed\n   record directory.\n   Verify that referenced live-role paths exist and expose competing aliases or\n   broken authority links.\n3. Read the current summary and search the compact history index for relevant\n   routes. Do not load a long dashboard, claims inventory or index in full just\n   to find the latest state. Open only the proof or research records needed to\n   verify conflicts or load-bearing claims.\n4. Do not choose a newer timestamp over stronger evidence. Expose unresolved\n   authority conflicts.\n5. Compare the live dashboard's review/checkpoint revision with later changes to\n   authoritative manuscripts, proofs and declared deliverables. A stale date is\n   a prompt to inspect, not by itself proof that the mathematics changed.\n\nIf `.mathbox/` is present, use the available `research-state` skill's brief\nread-only check and goal handoff, then inspect full details for affected claims,\nreview conditions and routes. Use `impact CLAIM` for a named claim's\ndependents and `pin-impact PATH` for the pins of a named file. Reconstruct\naffected proofs from artifacts;\ndo not merely repeat generated labels. Do not initialize or migrate state as a\nside effect of a read-only retrospective.\n\nDo not start a broad literature search merely to complete a retrospective. If\nthe requested review cannot be decided without establishing what a load-bearing\nexternal mathematical source says, route that bounded source question through\nthe available `literature-check` skill (`mathbox:literature-check` in plugin\ninstallations), which checks an authorized project-local cache before fetching.\nOtherwise record the unresolved source check as a candidate next route.\n\nIf `RESEARCH_LOG.md` still contains long-form legacy entries, read only the\nrelevant embedded entries and support a mixture of legacy prose and new links.\nReport the pending `mathbox:research-init` migration, but do not perform or\nrequire it as a precondition for the retrospective.\n\n## Build the portfolio\n\nFor each active claim or work package, record:\n\n- exact target and current evidence label;\n- durable evidence and review status;\n- load-bearing dependencies;\n- first unresolved implication or smallest counterexample;\n- recent route and why it succeeded or stopped;\n- expected scientific value, cost, and risk;\n- whether it lies on the current critical path.\n\nIdentify duplicated efforts, stale claims, abandoned routes with reusable\ninformation, and mutable facts incorrectly embedded in instructions.\nFor external or long computations, distinguish the last observed process state\nfrom current state. A launch record without a live process, scheduler result or\nlater observation is `unknown`, not `running`.\n\nGroup failed routes by their first failed mechanism rather than title. Identify\nshared unresolved dependencies and what mathematical change would reopen each\nroute. Distinguish new evidence from more prose, repeated bounded cases, and\nrediscovery of already recorded obstructions.\n\n## Select next routes\n\nRecommend at most three bounded research routes. Each must include:\n\n- exact unresolved mathematical question;\n- why it dominates nearby alternatives;\n- when current evidence is bounded, the uniform route or obstruction it\n  suggests;\n- cheapest decisive test and, when bounded, the alternatives it distinguishes;\n- success and failure criteria;\n- expected durable output;\n- dependencies and resource needs;\n- stopping condition.\n\nBalance one high-leverage route with lower-risk publishable or computational\nwork when the project permits. Do not keep a deliverable hostage to an unrelated\nopen flagship problem.\n\n## Review the AI workflow\n\nNote recurring guidance failures, false `mathbox` plugin skill triggers,\ncontext sinks, duplicated records, non-reproducible computations, or\nverification gaps. General plugin-skill bugs belong in the `mathbox` feedback\nledger; project rules belong in the repository.\n\nTreat a live status dashboard as current state, not verification history. Flag\nstacked dated verification narratives as a context sink. When reconciliation\nedits are requested, keep the latest full current summary and replace older\nnarratives with links to immutable research records or computation manifests.\nDo not rewrite indexed records or their history-index entries; append a linked\ncorrection record when history itself needs correction.\n\nFlag broken links to purported live dashboards, conflicts between the designated\nauthority and existing files, and completed deliverables still described as\nunresolved. Do not repair these during a read-only retrospective; identify the\nminimal reconciliation set.\n\n## Output\n\nLead with a concise project verdict. Then provide:\n\n1. claim/work-package table;\n2. contradictions or stale records;\n3. critical path and principal blocker;\n4. recommended routes in priority order;\n5. files to reconcile, only if edits were requested;\n6. the single best next prompt for the `mathbox:research-attempt` plugin skill.\n"
}