メニュー
Append a well-formed run summary to logs/agent-history.md after every agent pass. Shared by all agents.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Academic research assistant for literature reviews, paper analysis, and scholarly writing. Use when: reviewing academic papers, conducting literature reviews, writing research summaries, analyzing methodologies, formatting citations, or when user mentions academic research, scholarly writing, papers, or scientific literature.
Adversarial review of mathematical proofs, theorems, and formal arguments. Use when asked to review a proof, check a paper's correctness, verify a theorem, find bugs in mathematical reasoning, audit a derivation, or stress-test a formal argument. Implements iterative self-correction from Woodruff et al. (2026) and the Generator-Verifier-Reviser pattern from Aletheia (Feng et al., 2026).
Read, update, and promote claim statuses through the controlled vocabulary. Includes the project.yaml patch protocol and the conjectures.md update procedure.
Systematically test mathematical conjectures by searching for counterexamples, verifying small cases, and constructing proofs or refutations. Use when Raman proposes a conjecture, when a paper claims an unproven bound, when testing whether a theorem generalizes, or when asked to "find a counterexample," "test this conjecture," "verify this bound," or "check if this holds." Implements the simulation and counterexample search techniques from Woodruff et al. (2026) and the neuro-symbolic verification loop from Aletheia (Feng et al., 2026).
Generate research ideas by finding cross-disciplinary connections, retrieving obscure theorems from distant fields, and identifying analogies between different mathematical domains. Use when Raman asks for new research directions, when ingesting a paper that might connect to other fields, when looking for "bridges" between topics, or when asked to "find connections," "what other fields use this technique," or "cross-pollinate." Implements the cross-pollination techniques from Woodruff et al. (2026, Section 2.2, 4.x) and AI Behavioral Science's framework for studying AI-human strategic interactions (Jackson et al., 2025).
Classify each prose claim before writing to paper/paper.md. Enforce the controlled vocabulary and reject illegal proof language.
| name | run-summary |
| description | Append a well-formed run summary to logs/agent-history.md after every agent pass. Shared by all agents. |
Every agent pass must end by appending a structured entry to
projects/active/$SLUG/logs/agent-history.md. Use this skill to produce
a correctly formatted entry. Do not skip this step — the Supervisor reads
the log to determine next actions.
## <agent>: <task> (<ISO timestamp>)
### What changed
- <Bullet per artifact touched, e.g. "Appended 3-cluster section to `literature/map.md`">
### Evidence added
- <Bullet per citation, proof step, or data point added. For literature: paper IDs. For proofs: claim IDs and their new status. For experiments: exp-id and key metric.>
### Uncertainties
- <Anything you could not verify, marked [unverified], or left as an open question.>
### Next actions
- <Concrete recommended next step, e.g. "Run ideation pass on cluster 2" or "Attack C1 with induction">
Approval needed: yes|no
<agent> — Use the canonical role name, lowercase:
supervisor, literature, ideation, modeling, proof,
simulation, writer, verifier.
<task> — Brief imperative phrase matching the command that triggered
the run, e.g. literature review — influence maximization,
proof pass — C1, C3, update draft — section 4.
<ISO timestamp> — YYYY-MM-DDTHH:MM (no seconds needed). Use the
current date from the environment.
What changed — One bullet per file written or appended. Use backtick paths relative to the project root. Do not list files you only read.
Evidence added — Concrete and linkable. For literature: [Author et al., Year].
For proofs: C1: hypothesis → proof-sketch. For experiments: exp-003: fig-02 generated. Empty bullets are not acceptable.
Uncertainties — Items you marked [unverified] in the artifacts, cases
where the quality contract was only partially met, or open questions the
next agent should know about.
Next actions — At least one concrete suggestion. Not "continue as needed" — name the specific command and target.
If venue targeting is active, the first bullet under Next actions
should name:
target_venuemeets bar, close-needs-model, etc.)literature-review, refine-model,
proof-pass, experiment-pass, or retarget venue)Approval needed — yes if the run changed a claim to proved,
proposed a model-version bump, touched mode in project.yaml, or
involved any human-in-the-loop action. Otherwise no.
For compound playbooks (ideate, full-pipeline), append one entry that lists all sub-agents under "What changed":
## ideate: influence maximization (2026-04-11T14:30)
### What changed
- [literature] Appended 4-cluster section to `literature/map.md`.
- [ideation] Appended 5 candidates to `notes/idea_funnel.md`. Wrote 4 gap briefs.
- [verifier] Appended verifier pass to `reviews/suggestions.md`. Demoted 1 candidate.
- [writer] Wrote `reports/ideate-influence-maximization-2026-04-11.md`.
...
logs/agent-history.md is append-only. If the file does not exist yet,
create it with a single top-level heading:
# Agent history — <$SLUG>
…then append the first run entry below it.