Autoresearch Paper

Autoresearch Paper helps AI researchers keep long-running paper projects honest: it makes agents prove the algorithm or experiment works before they start writing.

Status

• Current version: v0.20.1
• MVP-0 migration track: P1 Research Compiler, P2 Minimal Worker Adapter, P3 Experiment Receipt ledger, P4 Evidence Gate, P5 Recompile Loop, and the P6 Supervisory Controller with complete L0/L1/L2 Watchdog closure are implemented on the codex/mvp0-thin-loop feature branch. P1 provides an isolated Research IR v1 schema, deterministic semantic validator, strongest-Codex compiler prompt, proposal/critique/revision workflow, two mandatory interactive Human Critique/Human Approval stops, and content-addressed freeze receipts. P2 binds an owner-reviewed IR to one clean detached research worktree and one exact Claude Code/MiniMax session, then validates a closed task contract, closed JSON result, Git-visible change boundary, and immutable identity/usage receipt. P3 consumes terminal P2 turns in exact order, archives pre-execution inputs and result/evidence bytes, publishes a closed content-addressed Experiment Receipt, and appends the full receipt plus digest to a replayable hash-chained JSONL ledger. P4 validates a closed report from the evaluator frozen in Research IR, binds it to exact P3 evidence, archives the evaluator implementation, and publishes one replayable KEEP/PIVOT/STOP/RECOMPILE decision per Experiment Receipt. P5 freezes one eligible P4 prefix, publishes one evidence-bound failure analysis and continuation/recompile request, and compiles an explicitly scoped Research IR N+1 back into P1 human review. A later P1 freeze is bound to the complete parent IR → Gate → analysis → request → child IR lineage. P6 derives one transition at a time from replayed P1–P5 truth, supports a fresh non-MiniMax delegated review only for deterministically bounded execution changes, resumes a failed predecessor through the exact fixed Claude/MiniMax session, and binds an independent zero-model launchd L0, an exact-thread Codex App heartbeat L1, and identity-bound Worker heartbeats L2. P2 now keeps the repository's Draft 2020-12 result schema authoritative while projecting a declaration-free Draft-07-compatible schema only at the Claude Code --json-schema boundary. This fixes Claude Code 2.1.205 rejecting the 2020-12 metaschema before a MiniMax request is launched. ENGINEERING_ACCEPTANCE is test-only; a live freeze requires recorded owner critique and later explicit owner approval. This track does not import or extend the Legacy Harness runtime and does not change the released v0.20.1 product claim. Dashboard remains a read-only projection and paper writing is not yet migrated into MVP-0. P6 deterministic and bounded field acceptance does not claim 24h, 7×24, production readiness, SOTA, or full autonomous-paper completion. See mvp/README.md.
• Stability: Experimental; Codex Host switched (Codex Host 已切换) after bounded T032 field-lineage acceptance
• Tier coverage: arxiv (open) · conference (gated) · journal-q1 (gated)
• Direction: The v1 target reverses the host boundary: Codex owns bootstrap, top-level planning, strong review, and loop control; one physically separate, plan-bound Claude Code session supplies MiniMax M3 Worker turns. T030 now implements exact --session-id/--resume delivery, exclusive send leasing, an immutable UUID/policy binding, receipt-chain rollback checks, joint Worker/session crash reconciliation, immutable instruction/turn receipts, and explicit cache-usage observations. An identity-uncertain live process remains unresolved with the session BUSY; only proven termination or absence permits terminal receipt publication. T031 provides the installed closed-brief entry, authenticated activation, and a transactional runtime bootstrap that exercises L0 recovery and the non-due L1 command path before reporting READY. T032 then passed on a fresh installed task: Stage 1 reached canonical RECORDED, a real strongest-model review accepted the frozen continuation, and Stage 2 resumed the exact same Claude session as turn 2 with real L2 heartbeats and a terminal receipt. v0.20.0 therefore switches the product Host role to Codex. This bounded result is not Stage 2 scientific completion, SOTA, 24h or 7×24 stability, production readiness, or full production cutover. MiniMax M3 workers, authenticated lifecycle authority, evidence gates, typed patrol, owned cleanup, the launchd-backed durable state loop, evaluator admission, capsule-bound MiniMax/Codex production transport, replayed scientific acceptance, deterministic integrity-failure routing, and two-stage gated learning promotion, and bounded seven-fault/multi-session acceptance are implemented. The packaged claude-research-conformance-v1 workflow is a closed M1 conformance fixture: it journals operation IDs and verifies terminal evidence, but does not claim to be the production topic-to-paper trigger. The production loop now has external registration, tick leases, canonical revisions, fresh context capsules, and evaluator-eligibility blocking; bounded fault/restart evidence is complete. The scientific-figure path now binds source data, transformations, renderer identity, commands, outputs, hashes, a figure-stage-frozen expected set, an exact inventory, and output-bound human review before writing. v0.18 adds the compiled Research Ledger Dashboard: one explicitly selected plan, fresh read-only inspection, bounded bound-log and dossier views, loopback-only serving, local assets, and no browser lifecycle authority. New staged plans freeze a human-owned optimization contract and exactly one first-stage envelope. CP-01 uses the strongest allowed Codex reviewer, while the deterministic controller remains authoritative. Each candidate gets one logical isolated Gate decision; terminal MiniMax reports receive fresh non-M3 review before at most one next stage is compiled. v0.17 makes state/staged_research/v1/ the sole runtime truth; state/progress.json and state/research-dossier.md are rebuildable, non-authoritative views. Capacity v2 separates per-stage and global Worker limits from STAGE-REVIEW and named checkpoint capacity. An initial signed contract may pre-authorize exactly one named next stage, and the controller may cross only after the prior stage has a terminal decision, MiniMax report, and fresh strongest-policy review. Silence is never approval. Observation-only bootstrap stages have an explicit non-Gate terminal path: preflight freezes path/hash/symbol/line bindings, the shipped executable validator checks the promoted inventory, and the controller records no Gate-accepted or reusable evidence before the mandatory strong review. Terminal scientific report content remains MiniMax-authored; after the call, the controller adds only the otherwise unknowable role-visible provenance hash to the canonical report. v0.16.2 expands content-addressed review material into the actual CP-01 audit context, rejects underfunded multi-call policies, and uses a real Claude-to-first-work-unit run as field acceptance. It rejects unavailable Claude executables and non-UTF-8 sources before budget mutation and reserves dispatch/scientific capacity under one controller lock. v0.16.1 added typed negative frontier advice, conservative unknown-usage charging, fail-fast stage-review routing, and signed prospective capacity grants without refunding launched calls. v0.17 claims only a bounded stage-crossing capability and acceptance target: one first-stage-terminal to second-stage-Worker-start crossing. It does not claim second-stage completion, scientific success, 24h or 7×24 stability, production readiness, or full cutover. MAVIS is available only as explicit legacy compatibility. See skills/autoresearch-paper/references/claude-code-runtime.md, the design notes in docs/evolution/design-review-2026-06-26.md and forward-looking plans in docs/ROADMAP.md.
• Maintenance: issues and PRs welcome; major refactors land in feature branches first.

Table of Contents

1. Why
2. Features
3. Quick Start
4. Architecture
5. When To Use
6. Install
7. Dependencies
8. Usage
9. Workflow
10. Repository Layout
11. FAQ
12. Boundaries
13. Contributing
14. Tests
15. Changelog
16. Citation
17. License

Why

Long-running AI research runs often fail in the same way: agents explore for hours, lose the thread, retry the same weak direction, then write an honest paper about a near-zero contribution. Autoresearch Paper makes that failure visible and recoverable with evaluator freeze, research acceptance gates, heartbeat watchdogs, and manifest-driven cleanup.

Features

• Blocks writing until the research gate passes or the human owner waives it.
• Tracks failed directions so agents pivot structurally instead of repeating the same dead end.
• Separates runtime stalls from scientific no-improvement using typed failures.
• Requires a plan-bound, independently registered L0/L1/L2 assurance closure before any unattended durable Worker can start.
• Requires signed, expiring, replay-protected pause, resume, stop, waiver, worker cancellation, and cleanup actions.
• Removes only exact-path, token-bound, plan-owned ephemeral resources.
• Verifies paper packages with artifact-only reviewer checks, not producer self-claims.
• Dispatches schema-bounded MiniMax M3 workers through one plan-bound Claude Code session and reserves a frozen budget before Codex review boundaries.
• Registers a session-independent launchd trigger with exactly-one tick claims, rebuildable canonical state, and fresh hash-bound task capsules.
• Treats state/staged_research/v1/ as the only staged runtime authority and uses rebuild-staged-projections to rebuild state/progress.json plus state/research-dossier.md as disposable operator projections.
• Blocks unattended conference/journal autonomy until evaluator authority, replay, regression, immutable inputs, search space, and complexity policy pass executable admission; any identity drift revokes eligibility.
• Serves one plan through a compiled Research Ledger Dashboard without adding a second progress store, a Node.js runtime dependency, or browser controls.

Quick Start

/autoresearch-paper — turn a research brief into a gated paper pipeline.
/autoresearch-paper status — inspect canonical, scheduler, Worker, process, log, and watchdog state without mutation.
/autoresearch-paper dashboard — open the loopback-only Research Ledger for one selected plan.
/autoresearch-paper stop — exact-once runtime shutdown, then report exact residual resources.

The installed Codex entry accepts one closed JSON brief. It validates every required field, path, permission, and budget before atomically publishing the plan; invalid input leaves no partial plan directory:

cd ~/.agents/skills/autoresearch-paper
python3 references/scripts/harness-runtime.py prepare-codex-host-plan \
  --brief /absolute/path/closed-brief.json \
  --plan-dir /absolute/owned/plans/new-plan

The returned immutable planning request assigns exactly one initial stage to the strongest Codex Host policy. After one authenticated authorize_contract, activate-codex-host-plan binds the first-stage materials and generated dossier; CP-01 and bootstrap-host-runtime then close the review and L0/L1/L2 runtime lineage. See the installed claude-code-runtime.md for the exact activation command.

Architecture

The v1 target control plane is Codex, with a deterministic file-backed controller between both model runtimes and formal plan state. Claude Code is a physically separate MiniMax M3 execution host, not lifecycle authority. Legacy MAVIS resources are compatibility-only. The target flow is:

Codex Host (bootstrap · plan · strong review · loop control)
        │
        ▼
deterministic controller ── canonical staged/durable state ── evidence ledger
        │
        ├── L0 launchd health supervisor (health-only, zero model calls)
        ├── L1 launchd durable work trigger (leases + state advance)
        └── L2 controller Worker heartbeat receipts
                │
                └── fixed Claude Code session
                        └── MiniMax M3 Worker proposals

T030 covers the fixed-session transport foundation. T031 moves the installed entry, Harness bootstrap, and first-stage plan compilation to Codex; T032 has now passed the bounded field lineage from Stage 1 terminal → strongest review → automatic Stage 2 compilation → second Worker turn in the same Claude session. The evidence and claim boundary are recorded in docs/evolution/codex-host-t032-acceptance-2026-07-30.md. Dashboard remains a read-only local operations view; Codex App task visibility complements it but does not replace canonical state.

L0, L1, and L2 are bound by one immutable activation receipt but keep distinct scheduler/command identities. Legacy Mavis cron, plan-l0-guard.py, and last_seen.jsonl remain compatibility artifacts rather than activation proof.

For a prepared and authorized Codex-hosted plan, compose and exercise the runtime layers in one idempotent transaction:

python3 references/scripts/harness-runtime.py bootstrap-host-runtime \
  --plan-dir PLAN --graph PLAN/durable-plan.json \
  --interval-seconds 300 --health-interval-seconds 1800 \
  --worker-stale-seconds 7200 --frontier-stale-seconds 7200 \
  --heartbeat-stale-seconds 3600

The command does not merely write plists. It runs a non-due L1 probe, removes the exact L1 service and requires L0 to restore it without model calls, freezes the L2 contract, and commits one READY receipt only after all bindings pass.

Research gate (T6.1/T6.2): The controller binds evaluator, evidence, threshold, candidate, and measured verdict hashes. Bare PASS text is rejected. Conference and journal writing additionally require an applied CP-04 final evidence audit; waivers are signed human records.

Unattended autonomy gate: Conference and journal-q1 durable triggers cannot register or advance without a current evaluator-admission receipt. The controller revalidates authority, replay, regression, inputs, search space, complexity policy, and exact evaluator identity on every autonomy boundary.

Resource manifest: Every target-owned removable resource is recorded in resource_manifest.json with an exact path, ownership nonce, and scope. Cleanup requires an authenticated receipt and refuses shared or escaping paths.

Pause / resume: Both actions require a signed, expiring, replay-protected human record. The deterministic controller writes canonical receipts and keeps the durable plan state available across Claude sessions.

Bounded staged continuation: Capacity v2 keeps the active stage's Worker quota, the plan-global Worker allowance, terminal STAGE-REVIEW, and CP-01/CP-02/CP-04 (optionally CP-03) in separate non-transferable classes. A frontier top-up never increases Worker capacity. If the initial signed authorize_contract names exactly one next stage, advance-staged-research may derive one bound continuation receipt after terminal decision/report/review evidence and start exactly one Worker for that next stage. No response or operator silence counts as approval.

For the deeper plan structure, see skills/autoresearch-paper/SKILL.md.

When To Use

Use this when you have a research idea, a target venue, and enough material or infrastructure to define an evaluator. Do not use it for one-off drafts, blog posts, slide decks, or camera-ready submission automation.

Install

Primary install path:

npx skills add WdBlink/autoresearch-paper -g

Upgrade copied installations to v0.20.1 with a full bundle refresh so the runtime and response schema move together:

npx skills add WdBlink/autoresearch-paper -g --copy

Then verify local runtime dependencies from the installed skill directory:

scripts/setup.sh

Install the focused scientific figure capability at the audited upstream revision. Do not install the complete collection merely for this workflow:

gh skill install K-Dense-AI/scientific-agent-skills \
  scientific-visualization \
  --pin 70a0d595e54b8d92ca54f216d4315e0ab8c7d967 \
  --agent claude-code --scope user

# Repeat only when Codex does not share the Claude/user skill directory.
gh skill install K-Dense-AI/scientific-agent-skills \
  scientific-visualization \
  --pin 70a0d595e54b8d92ca54f216d4315e0ab8c7d967 \
  --agent codex --scope user

scientific-schematics is optional and proposal-only. If installed, it may draft method or architecture diagrams, but neither its generated image nor an AI quality score passes the repository-owned figure gate.

For a project-level install, omit -g:

npx skills add WdBlink/autoresearch-paper

Legacy Mavis Runtime Registration

This skill can be invoked from Agent Skills-compatible runtimes after npx install. The target worker and frontier-advisor path does not require MAVIS. Old watchdog agent, cron, hook, and cleanup fixtures require it only when the caller explicitly selects the legacy compatibility path.

If your Mavis build does not scan Agent Skills directories, register the installed source directory into your Mavis skill root as a symlink or copy. Keep one source of truth; avoid maintaining a stale manual copy.

Dependencies

scripts/setup.sh checks the runtime surface and blocks with repair instructions if anything required is missing.

Usage

/autoresearch-paper

Topic: energy-aware UAV swarm coverage under wind disturbance.
Target: ICRA 2027.
Materials: PDFs and simulator notes in a local folder.

The skill asks for missing fields, confirms the tier, shows a readable plan preview, and only starts the controller after an explicit "go".

During a run:

Workflow

Repository Layout

autoresearch-paper/
├── README.md
├── skills/
│   └── autoresearch-paper/
│       ├── SKILL.md
│       ├── scripts/
│       │   └── setup.sh
│       ├── dashboard/                 # pinned React/Vite build source
│       ├── assets/
│       │   ├── task-prompt-snippets.md
│       │   └── first-action-last-seen-hook.md
│       ├── references/
│       │   ├── goal-keywords.md
│       │   ├── tier-decision-tree.md
│       │   ├── plan-template-arxiv.md
│       │   ├── plan-template-conference.md
│       │   ├── plan-template-journal-q1.md
│       │   ├── task-prompt-snippets.md
│       │   ├── research-state-contract.md
│       │   ├── lifecycle-contract.md
│       │   ├── claude-code-runtime.md
│       │   ├── frontier-response.schema.json
│       │   ├── figure-artifact.schema.json
│       │   ├── figure-requirements.schema.json
│       │   ├── scientific-figure-pipeline.md
│       │   ├── watchdog-prompt-template.md
│       │   ├── first-action-last-seen.md
│       │   ├── reviewer-readiness-rubric.md
│       │   ├── bootstrap-watchdog.sh
│       │   ├── launchd/
│       │   ├── dashboard/             # precompiled runtime assets
│       │   └── scripts/
│       │       ├── dashboard_server.py
│       │       └── harness-runtime.py
│       └── tests/
└── docs/

FAQ

Q: Can I run it on Codex CLI or Claude Code without Mavis? A: Yes. Policy, bounded MiniMax M3 workers, CP-01–CP-04 Codex gates, authenticated lifecycle actions, the durable trigger/state loop, evaluator admission/verdicts, patrol, and owned cleanup all run without MAVIS. Pass --legacy-mavis only for an old compatibility fixture.

Q: The research gate rejected my run. Can I waive it? A: Yes, with an expiring HMAC-signed waive_acceptance record. A Markdown string cannot waive the gate. Negative-result waiver is arxiv-only.

Q: What if the watchdog keeps reporting stalls? A: Check last_seen.jsonl and the L0 corruption guard output. Common causes: a worker task exceeded the 30-minute ceiling, the plan dir moved, or the JSONL is being written outside the plan dir. The rescue daemon (references/scripts/plan-rescue-daemon.py) is the entry point for diagnosis.

Q: Does the skill write the final paper? A: It produces a structured paper draft and evidence bundle. It does not submit to venues, does not promise a camera-ready PDF, and does not replace human authorship of novel claims. See Boundaries.

Q: How is cleanup different from stop? A: stop changes controller status and runs a crash-safe shutdown that disables the plan's L0, L1, frontier retry, and identity-matching Worker processes. It does not grant aggregate deletion authority. cleanup removes only individually approved, owned resource generations. Neither deletes paper outputs.

Boundaries

The skill produces a structured paper draft and evidence bundle. It does not submit to venues, does not promise a camera-ready PDF, and does not replace human authorship of novel claims. If the topic has no measurable evaluator, the skill downgrades to arxiv or stops for clarification.

Contributing

Run scripts/setup.sh test before opening a pull request. Changes that touch watchdog, L0, cleanup, or research gate behavior should add or update runtime contract tests under tests/.

Tests

cd skills/autoresearch-paper
scripts/setup.sh test

The test path runs contract validation, exhaustive runtime unit/negative tests, and a complete no-MAVIS fake-Claude/fake-Codex integration flow.

Changelog

Per-version notes live in skills/autoresearch-paper/SKILL.md#versioning. Quick highlights:

• v0.20.1 — turns the generic real-brief bootstrap failures from plan040 and plan043 into deterministic Runtime checks: real-positive Worker conformance, exact source-inventory construction binding, lifecycle proof aligned with strong-review reality, Runtime-derived continuation authority, and plan-wide deadline/frontier boundary evidence. The failed plans remain immutable; a fresh plan must pass CP-01.
• v0.20.0 — switches the product Host role to Codex after the fresh installed plan039 T032 lineage completed Stage 1 RECORDED → accepted strongest review → automatic Stage 2 compile → same-session turn 2 with real L2 heartbeat and terminal receipt evidence. The frozen acceptance report rehashed 70 evidence objects with zero mismatch. This is not a 24h, 7×24, SOTA, production-readiness, or full-production-cutover claim.
• v0.19.4 — prevents oversized execution dependencies from overflowing the Codex turn-start character ceiling: their immutable path, SHA-256, and size remain bound while only bounded review evidence is embedded. A second exact character preflight releases capacity before transport on overflow. T032 field lineage remains pending, so Host cutover is still not claimed.
• v0.19.3 — fixes the first live T032 audit findings: exact Worker digest delegation survives persistence/promotion, PAUSED can no longer compile a continuation, CP-01 closes evaluator/Dashboard/durable execution authority, and frontier evidence is embedded once with an exact pre-send estimate. T032 field lineage remains pending, so Host cutover is still not claimed.
• v0.19.2 — makes automatic Stage 2 a reviewed continuation: the terminal strongest-model review binds the exact next envelope, preflight inputs, and Worker task contract before compilation and dispatch. CP-01 also carries the actual Runtime and initial authorization directly. T032 field lineage remains pending, so Host cutover is still not claimed.
• v0.19.1 — closes the first T032 CP-01 audit failure by directly binding every first-Worker executable dependency and a Runtime-used exact-byte/order conformance module. Read-only Workers now delegate proposal hashing to the trusted Host. T032 field lineage remains pending, so Host cutover is still not claimed.
• v0.19.0 — adds the installed Codex Host closed-brief entry, atomic plan preparation, one authenticated first-stage activation, a predeclared persistent Claude session UUID/policy, and a READY receipt that binds the complete entry → staged state → durable L0/L1/L2/Dashboard/cleanup lineage. T032 field evidence is still required before the product claims Host cutover.
• v0.18.0 — adds the compiled, loopback-only Research Ledger Dashboard over fresh inspect-plan-runtime results, including typed absence, live/stale/empty/partial/mismatch/stopped/error presentation, bounded bound-log access, rebuildable dossier viewing, restrictive CSP, and Python-only installed serving. It also adds typed provider-quota recovery and exact-once logical retry lineages for named checkpoints. Adds the Claude-native runtime assurance closure: independent launchd L0 and L1 services, controller-owned L2 Worker heartbeats, interval validation, activation/test receipts, and health-only L1 recovery with zero model calls. A separate least-authority trigger can resume a due initial CP-01 quota retry before L1 admission.
• v0.17.2 — closes the remaining real Plan021 CP-01 findings: validator schema versions require JSON integers rather than booleans, and a Worker cannot pre-author Controller-owned role-visible provenance. Frozen source inventory and stage-report conformance suites now contain twelve and ten cases respectively; the former proves the actual CLI receipt path. Plan021 and Plan022 remain immutable negative evidence; the field gate requires a fresh plan and fresh real review. A pre-authorization prepare-staged-research pass now closes the Plan024 hash/receipt loop before any owner signature is created. Observation preflight also records exact Runtime path/hash, byte identity, and the ten-case conformance result for the terminal report validator, closing Plan025's sole CP-01 finding. The Worker output contract also preserves exact source symbol/line_start bindings separately from the generic input manifest, closing the Plan026 field defect where a valid inventory was rejected before content validation.
• v0.17.1 — explicitly inactive observation-only evaluation profiles and a CP-01-frozen, conformance-tested terminal-report validator close the execution-contract gaps found by real GPT-5.6 CP-01 field review. The six-role terminal review sees candidate, decision, and validation evidence; only an exact strongest-model accept may start the authorized next Worker. Older three-role terminal-review packets fail closed and require a fresh versioned stage/review path.
• v0.17.0 — sole-authority staged state with rebuildable projections, capacity v2 class isolation, and one idempotent, explicitly pre-authorized stage crossing through the start of one next-stage Worker. The isolated Codex surface now disables multi-agent features without the obsolete agents.enabled=false override, and an unattended Claude controller must be explicitly pre-authorized rather than relying on auto permission prompts. Observation-only Stage 1 now has a deterministic non-Gate terminal route, exact source symbol/line selection, an executable receipt-producing validator, and independent plan-local/Runtime byte-identity evidence. Canonical stage reports also remove a real-call circular dependency by allowing the controller to inject only the completed Worker's role-visible hash into otherwise unchanged MiniMax-authored content. Legacy capacity v1 retains existing-plan lifecycle/replay compatibility but cannot use automatic stage crossing. No second-stage completion, scientific success, long-soak, production-readiness, or full-cutover claim is made.
• v0.16.2 — field-loop recovery for substantive CP-01 evidence closure, aggregate frontier-budget admission, isolated single-reviewer Codex transport, external read-only Worker inputs, disclosed closed evaluators, and a real first-work-unit acceptance artifact.
• v0.16.1 — field recovery for frontier response/usage classification, a measured ChatGPT reservation floor, signed append-only capacity top-ups, content-addressed stage material, and deterministic CP-01/STAGE-REVIEW routing. No launched-call refund or apply-time capsule synthesis is added.
• v0.16.0 — bounded rolling stages, exact role-visible-state records, non-fungible CP-01/02/04 capacity, separately budgeted Gate transport retries, accept/reject/escalate evidence, fresh terminal strong review, and figure-inventory freeze at the authorized figure-production stage.
• v0.15.0 — MiniMax M3 may draft the top-level research plan, but Codex gpt-5.6-sol at ultra must independently accept the exact CP-01 evidence bundle before any worker dispatch; reviewer identity and policy hash remain revalidated by downstream consumers. The author family is controller-declared provenance, not cryptographic model attestation.
• v0.14.1 — frontier preflight before budget reservation, strict response schema compatibility, Git-safe HTTPS-only Codex routing, durable transport event streaming, and conservative accounting for uncertain sends. See the CP-01 incident report.
• v0.14.0 — source-bound scientific figure manifests, offline path/hash validation, post-KEEP pre-writing gates, focused Scientific Visualization integration, and optional proposal-only Scientific Schematics.
• v0.13.0 — seven production fault scenarios, multi-session soak evidence, and claim gates bounded to the measured interval; no 24h/7×24 claim.
• v0.12.0 — audited episode memory, defect-versus-lapse diagnosis, replay/regression-gated proposal receipts, and human-only evaluator proposals.
• v0.11.0 — controller-owned evaluator snapshots, replayed scientific acceptance, and isolated goal/evaluator integrity failure routing.
• v0.10.0 — capsule-bound MiniMax dispatch and Codex request derivation, with controller-only exact-once durable result commits.
• v0.9.0 — launchd-backed durable trigger, generation-bound tick leases, canonical plan revisions, fresh context capsules, metadata-only Guardian, and executable evaluator admission for unattended conference/journal plans.
• v0.8.0 — Claude Code target cutover with authenticated human actions, hash-bound evaluator and CP-01–CP-04 gates, typed failures, target patrol and owned cleanup, plus no-MAVIS end-to-end conformance.
• v0.7.0 — CLI → tool migration. The legacy mavis agent|cron|session|hook|archive CLI subcommands are removed by the runtime; the skill is rewired to use the native mavis tool (agent/cron/session) and direct file writes for hooks (~/.mavis/hooks/...). Only mavis team plan ... remains a CLI (with the v0.7 rename abort → cancel). mavis communication send is marked deprecated.
• v0.6.0 — Agent Skills monorepo layout (npx skills add support), cleanup-script subcommand fix, full test bundle under tests/.
• v0.4.0 — Platform-portable daemon pattern (no Linux setsid dependency), producer discipline, model preload for in-process NN pipelines.
• v0.3.1 — V6 evidence-driven optimization rounds (verifier spot-check, 0% framing recipe, wide-table camera-ready fix).
• v0.3.0 — Rescue Layer (L0 guard, watchdog daemon, abort gate, workspace isolation) and three failure-mode FMs.
• v0.2.0 — Three-tier plan templates and heartbeat contract.

For the complete history, see the git log of skills/autoresearch-paper/SKILL.md.

Citation

If this skill contributed to a paper or research artifact, you can cite the release as:

@software{autoresearch_paper,
  title  = {Autoresearch Paper: A Research-First Brief-to-Paper Pipeline
            with Evaluator Freeze and Heartbeat Watchdog},
  author = {WdBlink},
  year   = {2026},
  url    = {https://github.com/WdBlink/autoresearch-paper},
  version = {0.20.1}
}

License

MIT

Forged with Skill Forge · Crafted with Readme Craft