status - SKILL.md Agent Skill

name: status description: Generate a comprehensive project status report for the Open-FARS research pipeline. Shows pipeline progress, experiment status, paper progress, degradations, and next actions. Suitable for reporting to supervisors. user_invocable: true

Open-FARS Status Report

Overview

Generate a structured, comprehensive status report for the current Open-FARS research project. The report is designed to be presentable to a supervisor/PI — concise, factual, with clear progress indicators.

Usage

/status                    # Full report, printed to terminal
/status email              # Full report + send via email-notify
/status 重点关注实验进度      # Full report with emphasis on user's focus area
/status email 给老板看的周报  # Email report with user context

Parsing rules

If arguments contain email → generate report AND send via email-notify
Remaining text after email keyword → user focus / additional context to highlight
If no email keyword, all text is user focus

Data Collection Steps

Collect ALL of the following before generating the report. Use parallel tool calls where possible.

Step 1: Pipeline State (registry.yaml)

Read .open-fars/meta/registry.yaml:

Direction and project names
Each stage status (S1-S7)
Paper title and target venue
Known degradations
Review history (rounds, verdicts, scores)

Step 2: Experiment Progress

For each active experiment, collect:

Running processes:

ps aux | grep -E "run_e[0-9]|chain|gemma" | grep -v grep

E1 eval cache progress (for each task):

for task in ioi sentiment arithmetic country_capital nli; do
  db=".open-fars/projects/{project}/experiments/results/e1_landscape/$task/eval_cache.db"
  sqlite3 "$db" "SELECT COUNT(*) FROM eval_cache" 2>/dev/null
done

Latest experiment result files:

find .open-fars/projects/{project}/experiments/results -name "*.json" -newer <last_report_date> | sort

GPU utilization:

nvidia-smi --query-gpu=index,utilization.gpu,memory.used,memory.total --format=csv,noheader

Step 3: Paper Status

Check .open-fars/projects/{project}/paper/:

Does main.tex exist? When was it last modified?
Count TODO markers: grep -c "TODO" paper/sections/*.tex
Page count (from compilation log or aux file)
Latest judge review score and verdict

Step 4: Code Health

cd .open-fars/projects/{project}/code && python -m pytest tests/ --tb=no -q 2>&1 | tail -5

Step 5: Key Results Summary

Read the latest experiment result JSON files for each completed experiment (E1-E4) and extract headline numbers:

E1: fitness distribution (mean, std, range), autocorrelation, FDC
E2: number of circuit families, clustering quality
E3: H2 test result (within vs between, p-value)
E4: SAE FDC, tautology test result

Report Format

Generate the report in the following structure. Use Chinese for section headers (the user's supervisor is Chinese-speaking), but keep technical terms in English.

# Open-FARS 项目进展报告

**项目**: {paper_title}
**目标会议**: {target_venue}
**日期**: {today}
**方向**: {direction description}

---

## 一、流水线总览

| 阶段 | 状态 | 完成度 | 备注 |
|------|------|--------|------|
| S1 文献调研 | ✅ 完成 | 90 篇 | — |
| S2 创意生成 | ✅ 完成 | 5 个创意 | idea-01 入选 |
| S3 实验计划 | ✅ 完成 | v1 | 6 实验 + 5 消融 |
| S4 代码实现 | ✅ 完成 | 294 tests | — |
| S5 实验执行 | 🔄 进行中 | XX% | {detail} |
| S6 论文撰写 | 🔄 进行中 | 初稿 | Judge {score}/10 |
| S7 修改完善 | ⏳ 待开始 | — | 待 full data |

## 二、实验进度

### 正在运行

| 实验 | 任务 | 进度 | 预计 |
|------|------|------|------|
| E1 景观特征化 | IOI | 84% | — |
| ... | ... | ... | — |

### 已完成结果

{Key numbers from completed experiments}

### 降级与修复

| 编号 | 问题 | 严重度 | 修复状态 |
|------|------|--------|---------|
| D1 | ... | critical | ✅ 已修复 |
| ... | ... | ... | ... |

## 三、论文进展

- **页数**: {N} 页
- **Judge 评分**: {score}/10 ({verdict})
- **主要问题**: {top 3 issues from judge}
- **TODO 标记**: {count} 个

## 四、计算资源

| 资源 | 状态 |
|------|------|
| GPU | {N}×{type}, {utilization} |
| API | {model}, {status} |

## 五、风险与阻塞

{List any blockers, risks, or items needing human decision}

## 六、下一步计划

1. {next action 1}
2. {next action 2}
3. {next action 3}

{If user_focus_query: "\n## 七、重点关注\n\n{Detailed discussion of user's focus area}"}

Report Persistence

Every report (regardless of whether email is specified) must be saved to disk:

Generate timestamp: TZ=Asia/Shanghai date +"%Y-%m-%d_%H%M"
Save to: .open-fars/projects/{project}/status/{timestamp}_status.md
Directory: .open-fars/projects/{project}/status/ (create if not exists)
Files are append-only — never delete or overwrite prior reports
After saving, print the file path to the user

This ensures a persistent history of all status reports for the project, useful for tracking progress over time and providing to supervisors.

Email Mode

If email is in the arguments, after saving the report to status/:

Send via email-notify:

python3 .claude/skills/email-notify/send-email.py \
  --subject "[Claude Code] Open-FARS 项目进展: {one-line summary}" \
  --body-file .open-fars/projects/{project}/status/{timestamp}_status.md

Important

Facts only: never inflate progress or hide problems — the user needs accurate info for their supervisor
Quantify everything: use numbers, percentages, counts — not vague words
Flag risks early: if something might block progress, say so explicitly
Compare to plan: always reference the original research plan when discussing progress
Degradation transparency: always include the degradation table, even if all are fixed