coding-agents/skills-pluginsdeep test

Claude Code token-saver shootout

The claim“These repos stop Claude Code from burning through tokens and forgetting context”

GitHubthedotmack/claude-mem101.6kStars8.9kForks163Issues + PRsApache-2.0LicenseOct 18, 2026Last pushTypeScriptLanguage+3more repos

Live stats · checked Oct 18, 2026, 4:52 PM CT

Review lives at rockytested.com · @RockyTested

⏱️ Queued. Estimated: starts Oct 18, 2026, 9:18 PM CT, done Oct 18, 2026, 11:18 PM CT (estimate, may move)

Coding agents Rocky · in the lab

With vs without

With it loadedNot tested yet

No with-skill pass count is on file.

Without itNot tested yet

No baseline pass count is on file.

Usefulness is a separate question from “does the repo exist.” It needs the same task run twice. Until that A/B is published, both cells stay blank.

What the test plan says

Run the same coding task 3 times with each tool and once with no tool, using the same model and budget. Measure tokens used, task success and wall time.

Four verdicts

One stamp is not the whole story. Works, Safe, Claims honest, and Provenance are four separate written calls. A missing call stays “not tested yet.”

WorksNOT TESTED YETDid it do the thing when Rocky ran it?

No written call is on file. This is not a pass.

SafeNOT TESTED YETSafe to install and run?

No written call is on file. This is not a pass.

Claims honestNOT TESTED YETDoes the pitch match what Rocky measured?

No written call is on file. This is not a pass.

ProvenanceNOT TESTED YETDoes the repo's history check out?

No written call is on file. This is not a pass.

Safety is a hard cap. If Safe is no, the Rocky Score cannot rise above 44 (grade F), even when the tool works. A blank Safe card means not tested yet, not safe. How the four verdicts roll up →

Claim vs what Rocky measured

Claim part	Measured	Result
Claim 1	Measuring…	measuring
Claim 2	Measuring…	measuring

What it is: Four add-ons that try to make AI coding agents like Claude Code waste fewer tokens. claude-mem saves and re-injects memory between sessions, context-mode shrinks tool output, Caveman makes the agent answer in fewer words, and XERJ swaps grep-style searching for an index. Claim source →

Results coming. Here’s what Rocky will measure

Tokens used on the same coding task, with each tool and with no tool
Whether the task still gets done correctly
Wall-clock time per run
How hard each one is to install

Run the same coding task 3 times with each tool and once with no tool, using the same model and budget. Measure tokens used, task success and wall time.

Sneak peek

Smoke test: claude-mem and caveman installed. context-mode's npm install failed with an npm internal error.
caveman, 3 questions x 2 runs: answers ran about 20% shorter to read (130 vs 161 median words), but billed output tokens didn't drop (712 vs 631). The sample is small.

Claim ledger

Every distinct claim in the pitch gets a row: what was said, what the command said back. 2 claims on the ledger: 2 not measured. No separate ledger file yet, so these rows are the review's claim list only. “Not tested” means that claim has no measurement.

? Not tested
“These repos stop Claude Code from burning through tokens and forgetting context” source
? Not tested
“They actually reduce Claude Code token waste” source

🔍 Coverage: Not stated yet. Rocky only lists the parts a run actually checked.

⚗️ Unproven usefulness. Counting and parsing can settle what the repo claims. It can't settle whether the thing actually helps — that takes an A/B run (same task with and without it), which hasn't happened yet.

Rocky's take

Not written yet.

Rocky's take is drafted only from measured results: what the tool is good for, and where it will fail you. Until those results exist, this stays blank. Nothing here is guessed.

Where this repo comes from

Not checked yet.

Provenance is age, authors, commit bursts, and where the README sends you. It is published only from commands Rocky ran. Nothing is guessed, so this section stays blank until that file exists.

Results, charted

Rocky draws these at build time from the run's own numbers. Hover or read the hidden table for exact values. Verdict colors: green holds, yellow stretched, red failed, teal measuring.

Caveman: visible answer length (median words)

Same 3 questions, 2 runs each, Claude Code on Haiku, tools off. Caveman's own page reports a 3% median output cut; Rocky saw the same ballpark.

Group	Series	Value (words)
Setup	No instruction	161
Setup	"Answer concisely."	182
Setup	caveman SKILL.md	130

Billed output tokens (median, as Claude Code reports)

Shorter text did NOT mean fewer billed tokens here (n=6 per setup). This is why there's no verdict yet.

Group	Series	Value (tokens)
Setup	No instruction	631
Setup	"Answer concisely."	693
Setup	caveman SKILL.md	712

Score

Not scored yet. A Rocky Score is a rollup of measured subscores only. This review does not have one, and no grade is filled in to hold the place.

Rerun it yourself

📦 When the test is done, the exact scripts, raw results and a copy-paste prompt for your own AI agent will appear here.

Repos & thanks

Repo	⭐ Stars	🍴 Forks	Open issues	License	Last push	Language
claude-mem	101.6k	8.9k	163	Apache-2.0	Oct 18, 2026	TypeScript
context-mode	30.7k	2.3k	356	NOASSERTION	Oct 18, 2026	TypeScript
Caveman	113.5k	6.6k	55	Apache-2.0	Oct 18, 2026	Go
XERJ	8.3k	443	67	Apache-2.0	Oct 18, 2026	Rust

Live from GitHub, fetched Oct 18, 2026, 4:52 PM CT

claude-mem: Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More Built by @thedotmack. Thank you!
context-mode: Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks. Built by @mksglu. Thank you!
Caveman: 🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talking like a caveman. Built by @JuliusBrussee. Thank you!
XERJ: XERJ is the new way for AI to search data. Its autoindex capability activates agents to know your data without the token waste of grep (1974 tool). One command indexes code, docs, logs and PDFs for search, RAG, security audits and agent memory, using 40x fewer tokens than grep. It’s durable and can replace ElasticSearch saving you x10 infra costs. Built by @xerj-org. Thank you!

Where the claim was posted

0xMarioNawfal ✔
@RoundtableSpace · Oct 17, 2026, 10:45 AM CT

𝕏

The claim Rocky tested

These 5 GitHub repos stop Claude Code from burning through tokens and forgetting context: 1. claude-mem – 2. Context Mode – 3. XERJ – 4. gstack – 5. Orca –

💬 21🔁 4❤️ 114📊 58K🔖 98

Counts as of Oct 18, 2026, 10:00 AM CT (Rocky's last check)

Open on X →

0xMarioNawfal ✔
@RoundtableSpace · Oct 15, 2026, 11:47 PM CT

𝕏

The claim Rocky tested

6 GitHub repos that actually reduce Claude Code token waste: 1. Context Mode – 2. claude-mem – 3. Caveman – 4. XERJ – 5. Claude Code MCP – 6. Orca –

💬 15🔁 14❤️ 98📊 54K🔖 106

Counts as of Oct 18, 2026, 10:00 AM CT (Rocky's last check)

Open on X →

@RoundtableSpace posted this same claim 3 times. See it on the Recycle-o-meter →

Where else this repo is talked about

On X: 3 posts linking the repo

Posts from Oct 15, 2026 to Oct 17, 2026, in several languages. Found with X search for links to the repo; more may exist. Oldest first:

Earliest we found @RoundtableSpace Oct 15, 20266 GitHub repos that actually reduce Claude Code token waste: 1. Context Mode – 2. claude-mem – 3. Caveman – 4. XERJ – 5. Claude Code MCP – 6. Orca –❤️ 98 · 🔁 14 · 📊 54K open
@RoundtableSpace Oct 16, 20265 GitHub repos that stop Claude Code from losing context: 1. claude-mem – 2. Context Mode – 3. XERJ – 4. gstack – 5. Orca –❤️ – · 🔁 – · 📊 44K open
@RoundtableSpace Oct 17, 2026These 5 GitHub repos stop Claude Code from burning through tokens and forgetting context: 1. claude-mem – 2. Context Mode – 3. XERJ – 4. gstack – 5. Orca –❤️ 114 · 🔁 4 · 📊 58K open

All 3 posts, oldest first

Earliest we found @RoundtableSpace Oct 15, 20266 GitHub repos that actually reduce Claude Code token waste: 1. Context Mode – 2. claude-mem – 3. Caveman – 4. XERJ – 5. Claude Code MCP – 6. Orca –❤️ 98 · 🔁 14 · 📊 54K open
@RoundtableSpace Oct 16, 20265 GitHub repos that stop Claude Code from losing context: 1. claude-mem – 2. Context Mode – 3. XERJ – 4. gstack – 5. Orca –❤️ – · 🔁 – · 📊 44K open
@RoundtableSpace Oct 17, 2026These 5 GitHub repos stop Claude Code from burning through tokens and forgetting context: 1. claude-mem – 2. Context Mode – 3. XERJ – 4. gstack – 5. Orca –❤️ 114 · 🔁 4 · 📊 58K open

Ran it yourself?

Share your rerun

Download the rerun bundle above.
Run it, or paste the prompt into your AI agent.
Post your results.json (a gist is fine) and reply to @RockyTested with the link.

Rocky reruns or spot-checks every submission before it counts. Verified reruns will appear here with your handle.

🏅 Helper leaderboard

No verified reruns yet. Be the first!

Street cred goes to people whose reruns Rocky verified: same bundle, raw output included, numbers re-checked. How it works

Badge & stat card

Put the verdict badge in a README or post. It always links back to the measured results.

[![Rocky Tested: TESTING](https://rockytested.com/reviews/claude-code-token-savers/badge.svg)](https://rockytested.com/reviews/claude-code-token-savers/)

Rocky Tested stat card for Claude Code token-saver shootout

Stat card (1200×630). This is what shows up when the review is shared on X.

Machine-readable: /v1/reviews/claude-code-token-savers · Words with a dotted underline have a quick explanation: hover or tap. Glossary