worklog
Capture the "why" behind your code by turning git deltas and session thoughts into high-signal work journals.
Free
Works with the AI Agents you already use
worklog
Example session with this skill installed
记一下这次 session。我们优化了 scaler 的归一化逻辑,解决了测试集泄漏问题,但 cairo 渲染中文还是乱码。标记为 leakage-audit 线程。
- Read your context and instructions
- Compiled the worklog
I've captured the session details. I verified the `src/scaler.py` changes against the git diff to ensure the normalization logic was moved inside the cross-validation fold. I also noted the unresolved Chinese font rendering issue in Cairo. ## Session Log: 2023-10-27 ```markdown --- date: 2023-10-27 threads: ["leakage-audit", "cairo-render"] --- # Session: 14:30-15:45 ### 关键决策 / 为什么 - 归一化逻辑从全局预处理移至训练折内 (fit_transform):之前直接对全量数据 fit 导致测试集分布特征泄漏,修正后 cross-subject AUC 从 0.92 回落至真实的 0.88 (commit: d9e2f1b in src/scaler.py)。 ### 做成了什么 - 重构了 `DataPreprocessor` 类以支持 lazy-fitting。 - 删除了 `scripts/pre_process_all.py` 废弃脚本。 ### 死路 / 已排除 - 尝试通过调整 `StandardScaler` 参数解决偏差,但根本原因是数据流序错误,调参无意义。 ### 卡点 / 困惑 - `cairosvg` 在 Docker 环境下渲染中文依然显示为豆腐块 (fallback boxes)。已确认 `fonts-noto` 已安装,怀疑是 cairo 的 `font-config` 路径映射问题。 ### 碰到的 open thread - **cairo-render**: 需要在本地环境复现 Docker 下的字体加载失败路径。 - **leakage-audit**: 需进一步检查 `src/feature_eng.py` 是否有类似的全局统计量泄漏。 ``` **Next steps** 1. Run the `threads` command to see if other tasks are blocked by the cairo issue. 2. Check the logs in the `~/.claude/journal/2023/` directory to confirm the entry. 3. Start the next session by addressing the `cairo-render` debugging.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop losing the mental context that `git commit` messages miss. This skill captures the "why" behind your code changes, the dead ends you explored, and the open threads that would otherwise vanish when a session ends. It bridges the gap between raw git history and your actual reasoning process. ## What it does - **Capture sessions** transforms raw git deltas and session notes into high-signal work log entries. - **Track open threads** aggregates persistent task labels across multiple days to identify stalled work. - **Weekly distillation** synthesizes daily entries into structured weekly progress reports and priorities. - **Instant notes** allows capturing fleeting thoughts mid-session without breaking your flow. - **Context preservation** uses session snapshots to accurately reflect changes even if the session has already closed. ## How it works 1. Trigger `/worklog:journal` to collect git metadata and session notes via the `collect` command. 2. Draft the entry focusing on reasoning, dead ends, and hurdles rather than just listing changed files. 3. Pipe the structured entry to the `append` command to safely write to the versioned journal directory. 4. Use `threads` or `rollup` arguments to review active workstreams or generate weekly summaries. ## Frameworks & tools Built with Python, utilizing a dedicated `journal.py` CLI for git integration and filesystem management. Works within standard Git repositories but supports non-git directories through manual entry. ## Why this beats prompting it yourself Standard LLM prompts often hallucinate changes or simply repeat git logs. This skill enforces a verification loop, integrates real-time git diffs, and provides a structured schema that ensures your long-term project history remains searchable and high-signal. ## Use cases - Documenting why a specific architectural decision was made over an alternative. - Tracking a subtle memory leak thread across multiple days of intermittent work. - Generating a weekly status report for stakeholders based on ground-truth logs. - Handing off a complex task by listing specific "next steps" and current blockers. ## Known limitations Requires Python 3 or `py` in the environment. All write operations must go through the `append` command to ensure frontmatter validity. Non-git changes require manual verification to avoid hallucinations.
Requires Python 3 environment. Git diffs only capture tracked files; manual verification needed for gitignored files. Must use the provided CLI for all filesystem writes.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 19 days ago
- Free with an Agensi account
Requires