feat(EmailingEssay): 道の存在を書き、自分宛の口を開ける（v1.2.3）

## 道の明示

台帳を作った際、「読む読まないの記載はせず」という方針を
「振り返り側の文書には触れない」と解釈した。結果
"The ledger is memory, too." は送信スキルの File Locations にしか
置かれず、振り返って書く側の目に入らない場所に在った。

道の存在を伏せることは中立ではない。知らなければ「読む/読まない」
という選択自体が成立しないので、伏せることは見えない形の
「読むな」に等しい。reflect/SKILL.md に What the Plugin Retains を
独立した節として立て、何が残っているかを事実として書いた。
主語はプラグイン側に置き、扱いは書き手に返してある。
Reflection Process には段を足していない。

## 自分宛の口

沈黙の日（送らないと決めた日）に自分用のメモを残す手段が無かった。
sent/ は送信したものしか持たないため、書かなかった理由も残らない。

新しい保存機構は作らず、send に --to-self を足した。送信すれば
既存のデコレータが自動で台帳に載せるので、慣習ではなく構造で
担保される。台帳の recipient が宛先を語るため、エッセイと
書き置きは区別できる。send_custom() ではなく send(to=...) を
通すのは、前者が宛先に関わらず既定の受信者を台帳へ書き、
台帳が宛先について嘘をつくためである。

## 文書の三箇所

essay_wait.log について「全ての操作が記録される」と書いてあったが、
そのログを書くのは wait 経路だけだった。schedule 登録は
claude -p を schtasks へ直接渡してラッパを生成せず、例外の
monthly last_day のランナーも print するのみでログを開かない。
つまり実運用の主経路である定期便は、最初から書かない側にいた。

reflect/SKILL.md・send-email/SKILL.md・SETUP.md の三箇所を
経路を名指しする形へ揃えた。とくに SETUP.md の障害調査手順は、
最終行 2026-07-30 の古いログへ読者を送り「送信が試みられていない」
と誤読させる位置にあったので、診断を台帳から始める順に変えた。

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: web-flow

.claude-plugin/marketplace.json

@@ -33,7 +33,7 @@
       "name": "EmailingEssay",
       "source": "./EmailingEssay",
       "description": "Autonomous essay writing and emailing from your AI",
-      "version": "1.2.2",
+      "version": "1.2.3",
       "author": {
         "name": "Weave @ EmailingEssay"
       },

EmailingEssay/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "EmailingEssay",
-  "version": "1.2.2",
+  "version": "1.2.3",
   "description": "Autonomous essay writing and emailing from your AI",
   "author": {
     "name": "Weave @ EmailingEssay"

EmailingEssay/CHANGELOG.md

@@ -5,6 +5,41 @@ All notable changes to EmailingEssay will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.2.3] - 2026-08-28
+
+A note one can leave for oneself, and the reflection skill told what the plugin keeps.
+No behavior change to essay delivery.
+
+### Added
+- **`send --to-self`**. `python main.py send "Subject" "Body" --to-self` sends to
+  `ESSAY_SENDER_EMAIL`, the AI's own address — a place to leave a note on a silent day.
+  `LedgerRecordingMail` records it like any other send, so nothing new stores anything; the
+  ledger's `recipient` field is what tells a self-addressed note from an essay. There is no
+  general `--to <address>`: the plugin writes to the two addresses it already knows
+- `skills/reflect/SKILL.md` → **What the Plugin Retains**: what the plugin holds on its own
+  account (`essay_ledger.jsonl`, the bodies under `sent/`, `essay_replies.jsonl`, and a
+  `--to-self` note) is now stated where the reflecting side reads. It states the fact; it does
+  not ask that the records be read
+
+### Fixed
+- `skills/reflect/SKILL.md`, Non-interactive Mode: "Exit silently (logged to `essay_wait.log`)"
+  was not true. Scheduled runs invoke `claude -p` directly and never reach the wrapper that
+  wrote that log, whose last line is 2026-07-30. The text now says what holds — the silence is
+  recorded nowhere
+- `skills/send-email/SKILL.md`, File Locations and Security Considerations: "Logs for
+  wait/schedule operations" and "All operations are logged to `essay_wait.log`" claimed a reach
+  the log never had. Only the waiter generated by `wait` writes it — launch, target reached,
+  return code; a registered `schedule` invokes `claude -p` directly and never reaches that
+  wrapper. The text now scopes the log to `wait` and points at the ledger
+  (`essay_ledger.jsonl` + `sent/`), where every send is recorded
+- `SETUP.md`, "Email Not Received": the third instance of the same claim, and the one that
+  misleads during an incident — it sent the reader to a log a registered `schedule` never
+  writes, whose stale timestamp reads as "nothing was sent". Diagnosis now starts at the
+  ledger, and the log's `wait`-only scope is stated where it could otherwise be misread
+
+### Verification
+ruff 0 / ruff format clean / mypy Success (75 files) / pytest 511 passed
+
 ## [1.2.2] - 2026-08-28
 
 Closes the duplicate hole in the retroactive import. No behavior change to sending.

EmailingEssay/SETUP.md

@@ -146,8 +146,10 @@ SMTPAuthenticationError: Username and Password not accepted
 
 **Solution**:
 1. Verify recipient email is correct
-2. Check `~/.claude/plugins/.emailingessay/essay_wait.log` for errors
-3. Wait and retry (Gmail has daily sending limits)
+2. Look for the send in the ledger (`essay_ledger.jsonl` + `sent/`) — every send is recorded there
+3. For a `wait` run, `essay_wait.log` also holds the return code; a registered `schedule` never
+   writes it, so an old timestamp there is not evidence that nothing was sent
+4. Wait and retry (Gmail has daily sending limits)
 
 ### IMAP Connection Fails (reply ingestion)
 

EmailingEssay/commands/essay.md

@@ -305,6 +305,20 @@ python main.py ledger import-legacy --dry-run  # Preview a retroactive import
 Only mail whose `In-Reply-To` matches a ledger entry **and** whose `From` matches
 `ESSAY_RECIPIENT_EMAIL` is taken in — the inbox is not searched across.
 
+### Notes to self
+
+On a day with no essay to send, a note can still be left — addressed to
+`ESSAY_SENDER_EMAIL` instead of the reader:
+
+```bash
+python main.py send "Subject" "Body" --to-self  # Send a note to the AI's own address
+```
+
+No separate store is involved: the note goes through the same send path, so the same
+ledger records it. Its `recipient` field holds the sender address, which is what tells
+notes apart from essays. Such a note is never taken in by `replies fetch` — its `From`
+is not `ESSAY_RECIPIENT_EMAIL`.
+
 For file locations, see `skills/send-email/SKILL.md` → **File Locations** section.
 
 ---

EmailingEssay/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "emailingessay"
-version = "1.2.2"
+version = "1.2.3"
 description = "EmailingEssay - AI essay delivery via email"
 requires-python = ">=3.10"
 license = {text = "MIT"}

EmailingEssay/skills/reflect/SKILL.md

@@ -14,6 +14,7 @@ Design principle: "Reflection first, sending second."
 - [Invocation](#invocation)
 - [Reflection Process](#reflection-process)
 - [Output](#output)
+- [What the Plugin Retains](#what-the-plugin-retains)
 - [Essay Elements](#essay-elements)
 
 ---
@@ -92,7 +93,23 @@ Output: **Chat display only** (no email)
 Output: **Email** (user is not present)
 
 - **Delivering**: Use `skills/send-email` to deliver
-- **Not Delivering**: Exit silently (logged to `essay_wait.log`)
+- **Not Delivering**: Exit silently. Nothing records the silence — scheduled runs invoke
+  `claude -p` directly and never reach the wrapper that wrote `essay_wait.log`. A `--to-self`
+  note, if one is written, lands in the ledger like any other send.
+
+---
+
+## What the Plugin Retains
+
+The plugin keeps these records on its own account; what becomes of them in a reflection
+stays with the writer.
+
+- `essay_ledger.jsonl` — one line per sent essay; the bodies stay in `sent/` as `YYYYMMDD_HHMM.md`
+- `essay_replies.jsonl` — replies that `python main.py replies fetch` has ingested
+- `python main.py send "Subject" "Body" --to-self` sends to `ESSAY_SENDER_EMAIL` and lands in the
+  same ledger; the `recipient` field tells a self-addressed note from an essay
+
+Paths: `skills/send-email/SKILL.md` → **File Locations**.
 
 ---
 

EmailingEssay/skills/send-email/SKILL.md

@@ -48,6 +48,7 @@ yagmail
 |-----------|-----|-------------|
 | test | `python main.py test` | Send test email |
 | send | `python main.py send "Subject" "Body"` | Send custom email |
+| send | `python main.py send "Subject" "Body" --to-self` | Send a note to the AI's own address (`ESSAY_SENDER_EMAIL`); the ledger records it under that address |
 | wait | `python main.py wait TIME [OPTIONS]` | One-time schedule |
 | schedule | `python main.py schedule FREQ TIME [OPTIONS]` | Recurring schedule |
 | replies | `python main.py replies fetch` | Ingest replies to sent essays |
@@ -69,7 +70,7 @@ For full options and examples, see `commands/essay.md` → **Command Structure**
 
 | File | Location | Description |
 |------|----------|-------------|
-| `essay_wait.log` | `~/.claude/plugins/.emailingessay/` | Logs for wait/schedule operations |
+| `essay_wait.log` | `~/.claude/plugins/.emailingessay/` | Written by `wait` runs only; a registered `schedule` never writes it |
 | `schedules.json` | `~/.claude/plugins/.emailingessay/` | Backup of registered schedules |
 | `active_waiters.json` | `~/.claude/plugins/.emailingessay/` | Active waiting process tracking |
 | `runners/` | `~/.claude/plugins/.emailingessay/runners/` | Monthly schedule runner scripts |
@@ -100,11 +101,11 @@ The `wait` and `schedule` features use `--dangerously-skip-permissions` when lau
 **Safeguards**:
 - No file modifications or system changes are made by the essay command
 - Essay content is sent only to the configured `ESSAY_RECIPIENT_EMAIL`
-- All operations are logged to `~/.claude/plugins/.emailingessay/essay_wait.log`
+- A `wait` run logs its launch, the target time being reached, and the return code to `~/.claude/plugins/.emailingessay/essay_wait.log`; a registered `schedule` invokes `claude -p` directly and never reaches that wrapper
 
 **Best practices**:
 - Keep `ESSAY_RECIPIENT_EMAIL` set to your own email address
-- Review logs periodically with `cat ~/.claude/plugins/.emailingessay/essay_wait.log`
+- Review what was actually sent in the ledger (`essay_ledger.jsonl` + `sent/`, see File Locations), which records every send; `essay_wait.log` covers `wait` runs only
 - Audit registered tasks with `python main.py schedule list`
 
 ### Ingested Replies Are Untrusted Input

EmailingEssay/skills/send-email/scripts/adapters/cli/handlers.py

@@ -12,6 +12,7 @@
 from collections.abc import Callable
 from typing import TYPE_CHECKING
 
+from domain.config import Config
 from usecases.factories import (
     create_import_legacy_usecase,
     create_ingest_replies_usecase,
@@ -45,9 +46,15 @@ def handle_test(args: Namespace) -> int:
 
 @validate_config
 def handle_send(args: Namespace) -> int:
-    """カスタムメール送信"""
+    """カスタムメール送信（--to-self なら自分宛の書き置き）"""
     mail = get_mail_adapter()
-    mail.send_custom(args.subject, args.body)
+    if args.to_self:
+        # 自分宛は send() を通す。send_custom() は宛先を問わず既定の受信者を
+        # 台帳へ書くため、そちらへ流すと台帳の recipient が宛先を語らなくなる。
+        sender = Config.load().email.sender
+        mail.send(to=sender, subject=args.subject, body=args.body)
+    else:
+        mail.send_custom(args.subject, args.body)
     return 0
 
 

EmailingEssay/skills/send-email/scripts/adapters/cli/parser.py

@@ -80,6 +80,7 @@ def create_parser() -> argparse.ArgumentParser:
 Examples:
   python main.py test                          # Send test email
   python main.py send "Subject" "Body"         # Send custom email
+  python main.py send "Note" "Body" --to-self  # Send a note to the AI's own address
   python main.py wait 09:30 -t "morning"       # Schedule one-time essay
   python main.py wait list                     # List active waiting processes
   python main.py replies fetch                 # Ingest replies from the inbox
@@ -109,6 +110,11 @@ def create_parser() -> argparse.ArgumentParser:
     )
     send_parser.add_argument("subject", help="Email subject (メールの件名)")
     send_parser.add_argument("body", help="Email body (メールの本文)")
+    send_parser.add_argument(
+        "--to-self",
+        action="store_true",
+        help="Send to the AI's own address (自分宛に送る — 書き置き用)",
+    )
 
     # -------------------------------------------------------------------------
     # wait コマンド