v0.5: Add Two grill-me related skills for paper writing

1. 添加了2个grill-me相关的skills
2. paper-grill-me用于研究开始前的创新点判别器，用于：思考算法之前，确定论文 idea 之前，只有工程需求、研究方向或模糊想法时，怀疑当前“创新点”只是增加模块或迁移已有方法时。主要用于帮助用户逐步形成可辩护的创新点
3. paper-grill-with-docs用于论文完成后的 claim–evidence调和器，仅在以下条件基本满足时运行：实验已经完成，论文已有完整或接近完整的草稿，有结果文件、表格、日志或审计结果，用户想逐条检查论文结论是否被证据支持。

v0.4.3 Update Agents.md

1. Mandates at least two direct sub-proxies under the root proxy; non-recursive; single-writer model.
2. Introduces strict routing requirements for Sol, Terra, and Luna, along with disclosure rules for cases where routing cannot be specified at runtime.
3. Removes conflicting clauses regarding legacy dual-proxy setups, simulated roles, and fixed concurrency limits.

v0.4.2 : add multi-source support to paper-style-calibration

Add native multi-source support to paper-style-calibration skill.
Multi-source mode allows aggregating style patterns from 2-5 reference papers,
automatically extracting common patterns and filtering source-specific content

v0.4.1 : fix paper-style-calibration pipeline

Redesign paper-style-calibration with simplified user interface and fixed execution model

v0.4:add paper-style-calibration & enhance paper-polish-human

Add paper-style-calibration skill and enhance paper-polish-human with external style guide support.

paper-style-calibration (NEW):

• Calibrate writing style based on external samples (user's papers, reference papers, style descriptions)
• Support 4 input formats: TeX > Markdown > Word > PDF (prefer editable formats)
• Source kind: own (full extraction) vs reference (stats only, strictest guardrails)
• Default safety: unspecified source_kind treated as reference (most restrictive)
• De-identification rules: replace method/dataset/metric/value/benchmark with placeholders
• Comprehensive guardrails: content/expression/structural/context-level prohibitions
• Plagiarism prevention: 7+ word N-gram check, structural similarity check

paper-polish-human (ENHANCED):

• Add external style guide reading capability
• Check for psmfiles/STYLE_GUIDE.md with safe_for_paper_polish_human flag
• Apply humanization within style guide constraints
• Keep Manuscript-Local Voice Calibration as fallback when no external guide

v0.3 Improve paper-writing goal autonomy

• Add a goal-mode preflight blocker scan before the paper-writing pipeline starts
• Distinguish true blockers from non-blocking assumptions
• Require consolidated user confirmation only for blockers before Phase 0
• Log non-blocking uncertainty to AUTONOMY_LOG and continue execution
• Add goal-preflight reference with blocker taxonomy and report template
• Update paper-writing goal templates and final reporting expectations
• Sync installed paper-writing skill with repo version

v0.2: add paper polish human skills

Revise the paper by emulating human writing styles to avoid an overly polished or artificial tone; simultaneously, implement targeted adjustments to address specific issues commonly associated with revisions generated by current Large Language Models (LLMs).

v0.1

Rebuild the original repository from skills related to academic writing to skills covering the entire thesis process, including 7 modules and 26 skills