calez
Fumi / Email Log
This AI converts saved email threads (.eml / .mbox / pasted text) into an action log. It does not generate summaries or draft replies. Instead, it lists confirmed items, who took on what tasks and by when, pending requests that have not been accepted, conflicting deadlines for the same deliverable, and unanswered questions—all with direct quotes from the original text and message numbers. As a process specific to email, it automatically removes history quoted in replies. Even if the same commitment is quoted in four replies, it is counted as a single commitment. The number of lines removed is also clearly noted in the log. Forwarded requests are retained as part of the body text rather than as history. This is because deleting the forward would cause the request itself to disappear. The output consists of six fixed sections: Decisions, Agreements, Pending Requests, Inconsistencies, Unanswered Questions, and Extraction Notes. Each item includes a quote from the original text and a message number; the quotes are cross-referenced with the original text using the included verification script. Items with non-existent quotes are deleted. The only requirement is Claude Code (or Codex). Python 3.9 or higher is optional; if it is not available, “UNVERIFIED QUOTES” will be displayed in the log. There are no external APIs, account registrations, or communication during execution. It does not connect to IMAP, Gmail, or Outlook; it reads only saved files. There is no email sending functionality. What it cannot do: compose replies, summarize content, read attachments, merge multiple threads, analyze sent emails, or determine the assigned person. CC recipients are not treated as the assigned person. Requests remain as “pending requests” until a reply confirming acceptance is received. Supported languages are Japanese and English. For Japanese threads, the log is created in Japanese, and quotes are not translated. Since analysis and quote matching are performed on the Python side, the log format is preserved even if the model changes. Since extraction accuracy depends on the model, you can verify operation during setup using the included samples and the `tests/golden.md` file.