save-conversation

save-conversation

Save a concise summary of the current conversation to a markdown file in the conv-logs/ directory.

The goal is to create a readable record of what Toby and Claude discussed and accomplished — not a verbatim transcript, but a structured summary that's useful for future reference.

Workflow

IMPORTANT: Always resolve the project root via git rev-parse --show-toplevel and use that as the base for conv-logs/. Do NOT use the current working directory — it may be a subdirectory. If the caller is not inside a git repo at all, fall back to pwd so writes never escape into /conv-logs/....

Step 1: Create the output directory

Use the current date to create a hierarchical directory structure:

PROJECT_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
YYYYMM=$(date +%Y%m)
DD=$(date +%d)
mkdir -p "${PROJECT_ROOT}/conv-logs/${YYYYMM}/${DD}"

Step 2: Check for previous save in this session

Look for the most recent file across all subdirectories of conv-logs/ under the project root. Use a find -printf (Linux) / stat (macOS) sort instead of xargs ls -t so filenames with spaces or newlines stay safe:

LATEST=""
LATEST_MTIME=0
while IFS= read -r -d '' f; do
  if [[ "$OSTYPE" == "darwin"* ]]; then m=$(stat -f %m "$f"); else m=$(stat -c %Y "$f"); fi
  if (( m > LATEST_MTIME )); then LATEST_MTIME=$m; LATEST="$f"; fi
done < <(find "${PROJECT_ROOT}/conv-logs" -name 'conv-*.md' -type f -print0 2>/dev/null)
echo "$LATEST"

If a previous save exists, read it to determine where the last save ended — the new save should only include conversation that happened after that point. Always start numbering from ## 1. in each save file.

If no previous save exists, this is the first save of the session — include everything.

Step 3: Generate the filename

Use the current timestamp:

TIMESTAMP=$(date +%Y%m%d-%H%M%S)
YYYYMM=$(date +%Y%m)
DD=$(date +%d)
FILENAME="${PROJECT_ROOT}/conv-logs/${YYYYMM}/${DD}/conv-${TIMESTAMP}.md"

Step 4: Write the conversation summary

Review the conversation history (only the part after the last save, or everything if first save) and write a markdown file with this structure:

# Conversation Log — {YYYY-MM-DD HH:MM}

# 참여자
- Toby (사용자)
- Claude ({model name, e.g. Opus 4.6})

## 브랜치
- `{branch}` (or `main` if no feature branch)

---

## 1. {Topic Title}
사용: {comma-separated list of skills, tools, agents used in this exchange}

Toby: {One sentence summary of request}

Claude: {What was done — actions, key results. Use bullet list for multiple items:}
- item 1
- item 2

---

## 2. {Next Topic}
사용: {skills, tools, agents}

Toby: ...

Claude: ...

(Continue numbering for each distinct topic. Use `---` between sections.)

---

## 변경된 파일
- `path/to/new-file.md` (추가)
- `path/to/modified-file.json` (수정)
- `path/to/deleted-file.md` (삭제)

Before writing the file, run git diff --name-status HEAD (or compare against the commit at session start) to get the actual list of added/modified/deleted files. Use the status codes:

• A → (추가)
• M → (수정)
• D → (삭제)
• R → (이름변경)

If there are no uncommitted changes (everything was already committed and pushed), use git diff --name-status {first_commit_of_session}..HEAD to capture all changes made during the session. For incremental saves, only list files changed since the previous save.

Guidelines for writing the summary:

• Number each topic section sequentially (## 1., ## 2., ...)
• Give each section a descriptive title that captures the topic
• Toby: lines capture the user's intent in one sentence (not verbatim)
• Claude: lines describe what was actually done — use bullet lists when there are multiple actions or results
• Do NOT use bold markup (**) on Toby/Claude names — keep them plain text
• Add a 사용: line right after each section title listing skills, tools, and agents used (e.g., 사용: skill-creator, firecrawl, Bash, Explore agent)
• Group related back-and-forth exchanges into a single section under one topic
• Include file paths, command names, surface refs, and version numbers — these are the details that matter
• Write in the same language as the conversation (Korean if Korean, English if English)
• Skip trivial confirmations ("ok", "yes") — only log meaningful exchanges
• For incremental saves: always start numbering from ## 1. (do not continue from the previous file's numbering). Do not include a reference to the previous log file

Step 5: Save prompt rewrites

Review the conversation for any Korean prompts that were rewritten into English (lines like Your prompt rewritten: "..."). If any exist, append them to a monthly prompt log file:

PROMPT_FILE="${PROJECT_ROOT}/conv-logs/${YYYYMM}/prompt-${YYYYMM}.md"

If the file does not exist, create it with a header:

# Prompt Rewrites — {YYYY-MM}

| id | datetime | korean | english |
|----|----------|--------|---------|

Append each rewrite as a new row. Use an auto-incrementing id (continue from the last id in the file, or start at 1). The datetime is the approximate time of the exchange.

Example rows:

| 1 | 2026-03-25 21:30 | .idea는 .gitignore에 포함해줘 | Add .idea to .gitignore. |
| 2 | 2026-03-25 21:32 | save-conversation skill에서 md 문서를 만든 뒤에 git add로 tracking까지 해주도록 변경해줘 | Modify the save-conversation skill so that after creating the markdown file, it also runs `git add` to track the file. |

If there are no Korean-to-English rewrites in the conversation (or the scope being saved), skip this step.

After writing, run git add on the prompt file:

git add "${PROMPT_FILE}"

Step 6: Save and confirm

Write the conversation log file using the Write tool, then run git add to track it:

git add "${FILENAME}"

Then tell the user:

"대화 내용을 저장했습니다: conv-logs/{yyyymm}/{dd}/conv-{timestamp}.md"

Show a brief preview (first 10-15 lines) so the user can verify the content looks right.