Token Saver (English) — Cut Your AI Context Cost
Audit and reduce token consumption by slimming standing instructions and throttling runtime file reads.
$5
Works with the AI Agents you already use
Token Saver (English) — Cut Your AI Context Cost
Example session with this skill installed
My CLAUDE.md is getting huge and the agent is starting to ignore my project rules. Can you audit the standing context and suggest how to slim it down?
- Read your context and instructions
- Compiled the token saver english
I have audited your project instructions using the token_meter.py script against the Claude Code guidance.
Standing-context audit
Current harness: Claude Code
Budget: 200 lines (Guideline)
| File | Bytes | Est. Tokens | Current Lines | Risk |
|---|---|---|---|---|
| CLAUDE.md | 8,420 | 2,105 | 245 | High (Truncation) |
| AGENTS.local.md | 1,200 | 300 | 35 | Normal |
Expected after slimming: 8,420 -> 3,150 bytes (estimated)
Edit proposals awaiting confirmation
-
CLAUDE.md lines 110-165: Move the "Legacy API Documentation" section to a new file
docs/legacy_api.md.- Expected saving: 2,800 bytes / 55 lines.
- Reason: This content is rarely accessed and crowds out active coding rules.
-
CLAUDE.md lines 12-40: Consolidate redundant style rules.
- Current: 5 separate bullets for indentation and spacing.
- Suggested: "Follow Prettier config; use 2-space indents, no semicolons."
- Expected saving: 450 bytes.
-
CLAUDE.md lines 180-210: Move troubleshooting tables to
references/errors.md.- Expected saving: 2,020 bytes / 30 lines.
Next steps
- Provide a
yesto apply the migration of legacy docs todocs/legacy_api.md. - Confirm if I should proceed with the phrasing compression for style rules.
- Run
python3 scripts/token_meter.py doctorto ensure measurement precision.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
Large project instruction files like CLAUDE.md or AGENTS.md grow until they hit context limits or crowd out working memory. Coding agents often waste tokens by reading entire files when only a small range is needed.
What it does
- Audits standing context files against specific harness budgets for Codex, DeepSeek, and Claude Code.
- Generates diff proposals to slim down instructions by removing duplication and moving details to reference files.
- Throttles runtime token usage by using Grep to locate specific lines before reading large files.
- Intercepts large command outputs by redirecting to disk and slicing results instead of flooding the chat.
- Provides a Python-based token meter to accurately measure usage or provide weighted estimates when tokenizers are unavailable.
Frameworks & tools
Works with Claude Code, DeepSeek Harness, Codex, and WorkBuddy. Uses Python 3.10 standard library, tiktoken for precision, and shell utilities like grep, sed, and jq.
Why this beats prompting it yourself
Standard agents naturally default to reading full files and dumping long logs, which rapidly exhausts context windows. This skill enforces a "slice-first" strategy and budget-aware auditing that prevents silent truncation of critical instructions.
Use cases
- Compressing a bloated CLAUDE.md that has exceeded the 200-line recommended limit.
- Reducing API costs by preventing the agent from re-reading files already in the session.
- Auditing project instructions to see which files are at risk of being ignored by the harness.
- Managing long-running test suites by capturing logs to disk and only surfacing error traces.
Known limitations
Never modifies session history or instruction files directly. Edits require human confirmation. Measurement accuracy drops for CJK characters if tiktoken is not installed.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Requires manual confirmation for all file edits. Measurement is an estimate if tiktoken is missing. Cannot modify session history files.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 months ago
- Software
Python