Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add yanfeid/claude-amber --skill dietgit clone --depth 1 https://github.com/yanfeid/claude-amberWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/yanfeid/claude-amber/diet)<a href="https://agentmods.dev/skills/yanfeid/claude-amber/diet"><img src="https://agentmods.dev/badge/skills/yanfeid/claude-amber/diet/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/yanfeid/claude-amber/diet"><img src="https://agentmods.dev/badge/skills/yanfeid/claude-amber/diet.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00061 | $0.00581 |
| Opus 5 | $0.00030 | $0.00291 |
| Sonnet 5 | $0.00012 | $0.00116 |
| Haiku 4.5 | $0.00006 | $0.00058 |
Grade A, and why
diet scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 37 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/amber:diet — context audit & slimming plan
Audit this project's fixed context costs and produce a slimming plan. Fixed costs are paid every turn of every session — a 2,000-token CLAUDE.md in a 100-turn session is 200K tokens of spend and a fatter, dumber context.
1. Measure the consumers
Check each of these (skip silently if absent):
| Consumer | How to check | Red flag |
|---|---|---|
Project CLAUDE.md (+ imports) |
Read it, estimate tokens (~chars/3.5) | > 300 lines |
User ~/.claude/CLAUDE.md |
Read it | Project-specific content in a global file |
| MCP servers | .mcp.json, ~/.claude.json mcpServers for this project |
Servers with 10+ tools that are rarely used — every tool schema is always-loaded context |
.claude/rules/, large always-on context files |
List + size | Rules that only apply to specific situations |
.claude/skills/ |
List descriptions | Bloated descriptions (the description line is the only always-loaded part — bodies are free) |
2. Classify every instruction in CLAUDE.md
- Keep (always-relevant): build commands, hard constraints, style rules that apply to every edit.
- → Skill (situational): anything starting with "when X, do Y" that isn't hit every session — deploy procedures, release checklists, how to run a specific tool. Move the body into a skill; only its one-line description stays resident.
- → Delete: things Claude can derive from the code, duplicated docs, stale instructions.
3. Deliver the plan
Output a short report:
- Current fixed cost estimate (tokens per turn) by consumer.
- Specific moves: "lines X–Y of CLAUDE.md → new skill
deploy-checklist", "disable MCP server Z (used 0 times in recent sessions, costs ~N tokens of tool schemas)". - Estimated fixed-cost after the diet.
Then ask whether to apply the moves. If the user approves, perform them: create the skill files (with description frontmatter written for recall, ≤ 2 lines), trim CLAUDE.md, and list every change made. Do not delete anything without showing it first.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 37 lines · 61 tokens per session scan A b37f538d1c11
diet is a skill published in the GitHub repository yanfeid/claude-amber (2 stars, last pushed 2mo ago), licensed MIT. It adds 61 tokens to every session and 581 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
debug
Use when encountering any bug, test failure, unexpected behavior, or performance regression — before proposing any fix. Root-cause discipline plus a feedback-loop method; applies at every tier.
ship
Use when work is verified and ready to leave the machine — branch/PR/merge/deploy/smoke-check. Triggers on "ship it", "deploy", "merge the PR", "push this", "release", "go live", "cutover". Covers git hygiene, PR and merge discipline, deploy-target detection (Cloudflare Workers/Pages via wrangler, Vercel, Convex…
browser-qa
Use to exercise the running app in a real browser — after UI changes, before ship, or when the user asks to QA the site, smoke-test a branch, or check "does it actually work". Drives affected routes headlessly via Playwright, captures console errors, failed requests, and screenshots, and produces a report with a…
design
Use when designing or improving a module's interface, deciding where a seam goes, making code more testable or navigable, or pinning down domain terminology. Provides the deep-module vocabulary other skills lean on.
learn
Use at the end of T3 work, after an incident/rollback, or when the user asks for a retro — turns what happened into recorded learnings that live in the repo (CHANGELOG entry, typed LEARNINGS.md entries, follow-up tasks, and — only with explicit user approval — new RULES).
verify-done
Use when about to claim work is complete, fixed, or passing — before committing, opening a PR, or telling the user it works. Requires running verification commands and confirming output first; evidence before assertions, always.