Commands: save a routine you often run as /name
Command: a one-key routine
A Command is .claude/commands/name.md; the description at the top appears in the / menu. The kit has 3; full content below.
.claude/commands/estimate.md Download---
description: Estimate before starting: effort, risks, what the user must provide
---
When the user types /estimate <request>, do NOT write code yet. Output:
1. My understanding of the request (two or three sentences, for alignment).
2. Split into 3-7 steps; for each: expected difficulty (low / medium / high) and whether it can be verified automatically.
3. Dependencies: what the user must provide (keys, accounts, assets, decisions); write "none" if none.
4. The 2 biggest risks, and which step would expose them fastest.
5. One recommended order to start.
Rules: no time promises, only relative size; where unsure say "unsure" instead of inventing an exact-looking number.
.claude/commands/session-end.md Download---
description: End-of-session save: what was done, what is unfinished, update RESUME.md for next time
---
When the user types /session-end, do this in order:
1. List every file changed or added in this session (use `git status` or your own list; full paths).
2. Append a section to the end of `SESSION_LOG.md`: date / what was done / what is unfinished / pitfalls met.
3. Update `RESUME.md` so it only contains: where to start next time (3-5 items), decisions the user must make, key file paths.
4. If `reply.md` is over 2000 lines, tell the user to back it up; do not delete anything yourself.
5. Finally tell the user: which files were added/updated (full paths), and the first sentence to say to Claude next time.
Never put keys or passwords in the save notes.
.claude/commands/status.md Download---
description: Show the workspace state: req/reply line counts, open tasks, next step from last session
---
When the user types /status, output a short status report:
1. Count the total lines of `req.md` and `reply.md` (use `wc -l`, not methods that miscount) and find the line number of the last `══` block in reply.
2. Read `TASKS.md` (if present) and list "in progress" and "pending" items, high priority first.
3. Read `RESUME.md` (if present) and extract the "next step".
Output format:
```
=== Current status ===
📄 req: XX lines reply: XX lines (last block at line XX)
🔄 In progress: ...
📋 Pending (high priority): ...
📌 Next step from last session: ...
```
If a file does not exist write "(none)"; do not make things up.