| name | ashlr-lean-tools |
| description | Highest-impact token-saving habits for Claude Code tool use. Read-once, batch edits, orient-before-search, skip verification reads, grep-before-read, concise output at high spend. Reduces session token cost 30–60%. |
Ashlr Lean Tools
A discipline layer for tool use: each rule targets a specific token-waste pattern that commonly appears in long sessions.
Rules (active when this skill is enabled)
1. Read-once
Do not re-read a file that is already in context. If you opened src/auth.ts earlier this session and haven't been told it changed, use the content you already have. Re-reads are the single largest source of avoidable input-token spend.
2. Batch edits via ashlr__multi_edit
When making 2+ edits to the same file, collect them into a single ashlr__multi_edit call instead of sequential ashlr__edit calls. Multi-edit applies all changes in one round-trip and returns a single compact diff.
3. Orient before searching
Run ashlr__orient once at the start of each task. The genome overview typically resolves "where does X live?" without any file reads. Skip orient only when continuing work in a file you already have open.
4. Skip verification reads after a successful edit/write
ashlr__edit and ashlr__write confirm success by returning a diff. Do not read the file again just to confirm the change landed — that's redundant and costs the full file content in tokens.
5. Grep-before-read
Before opening a file to find something, run ashlr__grep first. If the grep result shows the relevant lines, use them directly — no full file read needed.
6. Shift to concise output when session spend is high
When the session has accumulated >5,000 tokens of tool-call output, prefer shorter responses: skip preamble, use file:line citations instead of quoted blocks, and omit trailing summaries when the diff speaks for itself.
7. Prefer ashlr__bash over native Bash for verbose commands
ashlr__bash auto-compresses long stdout (test runners, git log, find, npm ls) — 60–90% token savings on large output.
Anti-patterns
- Re-reading a file that hasn't changed since you last read it
- Two sequential
ashlr__edit calls on the same file (use ashlr__multi_edit)
- Opening a file just to confirm an edit you just made
- Running
Read before Grep to find a function
- Quoting large file contents in prose when a
file:line reference suffices
Activation
Enable via /ashlr-lean-tools on or by writing {"enabled":true} to ~/.ashlr/lean-tools.json. Toggle off with /ashlr-lean-tools off.
