scientify 1.13.3 → 1.13.4

package/README.md

@@ -377,6 +377,10 @@ Behavior notes:
 - Built-in quality gates use soft mode by default: non-fatal quality gaps are stored as warnings (`quality_gate.severity=warn`) while the run remains `ok`; only fatal issues are downgraded to `degraded_quality`.
 - Default strict recall behavior uses tiered retrieval (topic-related + broader method + adjacent cross-domain) to avoid sparse core selection on broad topics.
 - Each run executes at most one immediate reflection follow-up and writes trace/results back to `knowledge_state` for auditability.
+- Hypothesis-facing response policy is gate-driven:
+  - if `knowledge_state_summary.hypothesis_gate.accepted == 0`, Scientify returns factual cycle status only (no speculative roadmap/deep-dive suggestions)
+  - if `accepted > 0`, Scientify includes hypothesis details in the current message by default
+  - when channel/runtime clearly supports multi-send, it may optionally split into two consecutive pushes (brief alert + detailed hypothesis)
 - Lightweight preference memory is stored backend-only (keyword/source affinities) and used to rerank future pushes quietly.
 - Incremental dedupe + memory state is persisted under `~/.openclaw/workspace/scientify/` (`literature-state.json`, `literature-push-log.jsonl`).
 - Project-level research traceability is persisted under `~/.openclaw/workspace/projects/{project-id}/knowledge_state/`.

package/README.zh.md

@@ -374,7 +374,11 @@ openclaw gateway
 - 一次性研究任务（`at ... --topic ...`）使用“代表论文聚焦检索”；周期任务（`daily/weekly/every/cron`）保持“增量追踪”模式。
 - 周期增量模式会先构建候选池并评分后再选 Top-K；若本轮无“未推送新文献”，会自动再跑一轮代表性回退检索，再决定是否返回空结果。
 - 默认 `max_papers` 为 5（可用 `--max-papers` 覆盖）。
-- 研究记录内置质量闸门：核心论文全文覆盖率 >= 80%、证据绑定率 >= 90%、引用错误率 < 2%。未达标会自动降级为 `degraded_quality`。
+- 研究记录内置质量闸门：核心论文全文覆盖率 >= 80%、证据绑定率 >= 90%、引用错误率 < 2%。默认采用 soft gate：非致命缺口会记录为 warning（`quality_gate.severity=warn`）但保留 `ok`，仅 fatal 问题才降级为 `degraded_quality`。
+- 假设相关回复采用门禁驱动：
+  - 当 `knowledge_state_summary.hypothesis_gate.accepted == 0`：只返回事实型周期状态（论文/阅读状态/变更/门禁），不输出泛化“下一步路线/deep dive”建议。
+  - 当 `accepted > 0`：默认在当前消息内直接给出假设详情。
+  - 仅在渠道/运行时明确支持多次发送时，才可拆成两条连续推送（简报 + 详情）。
 - 轻量偏好记忆（关键词/来源亲和）仅后台保存，不默认展示给用户，会静默影响后续排序。
 - 增量去重与偏好状态会持久化到 `~/.openclaw/workspace/scientify/`（`literature-state.json`、`literature-push-log.jsonl`）。
 - 项目级研究可追溯状态会持久化到 `~/.openclaw/workspace/projects/{project-id}/knowledge_state/`。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "scientify",
-  "version": "1.13.3",
+  "version": "1.13.4",
   "description": "Scientify - AI-powered research workflow automation for OpenClaw. Includes idea generation, literature review, research pipeline skills, and arxiv tool.",
   "type": "module",
   "main": "dist/index.js",

package/skills/research-subscription/SKILL.md

@@ -110,6 +110,10 @@ For selected core papers, prefer full-text reading first:
   - do not emit BRIDGE unless `evidence_ids` resolve to this run's papers and include at least one full-text-backed paper
 - Hypothesis gate:
   - avoid speculative guesses; each hypothesis should include >=2 `evidence_ids`, `dependency_path` length >=2, and novelty/feasibility/impact scores
+  - before user-facing text, read `status.knowledge_state_summary.hypothesis_gate.accepted`
+    - if `accepted == 0`: output factual cycle report only (papers/read-status/changes/gates); do not output speculative "high-value routes"/"deep dive" guidance
+    - if `accepted > 0`: include hypothesis details in the current message by default (stable delivery path)
+    - only if runtime/channel clearly supports multi-send, optionally split into two consecutive messages (alert first, details second)
 If an incremental pass returns no unseen papers, run one fallback representative pass before returning empty.
 If user gives explicit preference feedback during follow-up (read/skip/star style intent, source preference, direction preference),
 persist it via `scientify_literature_state` action=`feedback` (backend-only memory, not user-facing by default).