Skip to content
Back to search

supercompress

MarketplaceWorks with Claude CodeGitHubIncomplete

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Supercompress/Supercompress with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Supercompress/Supercompress'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/supercompress--supercompress/badge.svg)](/listings/supercompress--supercompress)

HTML

<a href="/listings/supercompress--supercompress"><img src="/api/listings/supercompress--supercompress/badge.svg" alt="findsafeskills safety rating" /></a>

Install

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 33 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "name": "supercompress",
  3. "owner": {
  4. "name": "SuperCompress",
  5. "email": "team@supercompress.dev",
  6. "url": "https://www.supercompress.dev"
  7. },
  8. "metadata": {
  9. "description": "SuperCompress \u2014 query-aware context compression for coding agents.",
  10. "version": "1.0.0"
  11. },
  12. "plugins": [
  13. {
  14. "name": "supercompress",
  15. "source": "./integrations/plugins/supercompress",
  16. "description": "SuperCompress v2 \u2014 hosted ~400M Neural Keep. Compress tool dumps before they burn tokens. Hosted MCP supports OAuth 2.1 sign-in (PKCE). Plugin no-key setup is next. Free tier: 5M tokens/month, then $0.10/1M.",
  17. "version": "0.5.38",
  18. "author": {
  19. "name": "SuperCompress"
  20. },
  21. "homepage": "https://www.supercompress.dev",
  22. "license": "MIT",
  23. "category": "productivity",
  24. "keywords": [
  25. "context-compression",
  26. "prompt-compression",
  27. "token-compression",
  28. "llm-cost",
  29. "mcp"
  30. ]
  31. }
  32. ]
  33. }
README.md · README · 202 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <a href="https://www.supercompress.dev">
  3. <img src="https://www.supercompress.dev/assets/img/og-share-light.png" alt="SuperCompress — cut LLM context waste" width="840" />
  4. </a>
  5. </p>
  6. <h1 align="center">SuperCompress v2</h1>
  7. <p align="center">
  8. <strong>Query-aware context compression for AI applications and coding agents.</strong><br />
  9. Cut coding-agent context ~64% in our B5 benchmark.
  10. </p>
  11. <p align="center">
  12. <strong>~400M Neural Keep</strong> on the hosted API · coding-agent plugin via MCP<br />
  13. <a href="https://huggingface.co/arjunkshah21/sc-keep-crossencoder-v4-large">Open weights (MIT) on Hugging Face</a>
  14. </p>
  15. <p align="center">
  16. <strong>64.1%</strong> mean context cut on B5 (24 coding-agent cases) · <strong>24/24</strong> evidence passes · up to <strong>96.6%</strong> @ ≥99% retention<br />
  17. <em>(B5 coding-agent suite · evidence containment metric — not downstream LLM completion)</em>
  18. </p>
  19. <p align="center">
  20. <a href="https://www.supercompress.dev/dashboard?signup=1">Get API key</a> ·
  21. <a href="https://www.supercompress.dev/arena">Arena</a> ·
  22. <a href="https://www.npmjs.com/package/supercompress-proxy">Install</a> ·
  23. <a href="https://www.supercompress.dev/benchmarks">Benchmarks</a> ·
  24. <a href="https://docs.supercompress.dev">Docs</a>
  25. </p>
  26. <p align="center">
  27. <a href="https://www.supercompress.dev/supercompress-vs-headroom">vs Headroom</a> ·
  28. <a href="https://www.supercompress.dev/supercompress-vs-rtk">vs RTK</a> ·
  29. <a href="https://www.supercompress.dev/supercompress-vs-llmlingua">vs LLMLingua</a>
  30. </p>
  31. <p align="center">
  32. <a href="https://pypi.org/project/supercompress/"><img src="https://img.shields.io/pypi/v/supercompress?style=flat&logo=python&logoColor=white&label=PyPI" alt="PyPI" /></a>
  33. <a href="https://www.npmjs.com/package/supercompress-proxy"><img src="https://img.shields.io/npm/v/supercompress-proxy?style=flat&logo=npm&logoColor=white&label=npm" alt="npm" /></a>
  34. <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-3da639?style=flat" alt="MIT License" /></a>
  35. <a href="https://github.com/Supercompress/Supercompress"><img src="https://img.shields.io/badge/GitHub-Supercompress-181717?style=flat&logo=github&logoColor=white" alt="GitHub" /></a>
  36. <a href="https://www.supercompress.dev/dashboard?signup=1"><img src="https://img.shields.io/badge/Sponsor-EA4AAA?style=flat&logo=githubsponsors&logoColor=white" alt="Sponsor SuperCompress" /></a>
  37. </p>
  38. ---
  39. ## Why it exists
  40. Every LLM call ships a pile of context: RAG chunks, chat history, tool dumps, logs, JSON. Most of it is irrelevant to the *current* question — but the model still reads it, and you still pay for it.
  41. | Usual “fix” | What actually happens |
  42. |---|---|
  43. | **Truncate** | Deletes the middle. The answer is often in the middle. |
  44. | **Summarize** | Rewrites evidence. IDs, stack traces, and exact errors get soft. |
  45. | **Hope** | Ship the full dump. Watch the bill climb. |
  46. **SuperCompress v2** compresses context **against the query**. It keeps answer-critical lines in their original wording and drops the rest — on our coding-agent benchmark (B5), **64.1% mean context reduction** with **24/24 evidence-retention passes**.
  47. ---
  48. ## Product
  49. SuperCompress is a compression layer in front of inference:
  50. 1. Takes a long **context** + the current **query**
  51. 2. Segments and scores blocks by relevance to that query
  52. 3. Keeps entities, errors, definitions, nearby dependencies
  53. 4. Returns a smaller prompt + token stats
  54. The **query is never compressed** — only the surrounding context.
  55. ### Two paths
  56. | | **Neural v2 (recommended)** | **Compiler (fast/local)** |
  57. |---|---|---|
  58. | **Engine** | ~400M query-aware cross-encoder | Lightweight local policy |
  59. | **Runtime** | Hosted Fly CPU (Neural Keep) | Local CPU · millisecond-class |
  60. | **Best for** | Highest-quality keep on agent dumps | Local preprocessing / speed |
  61. | **Benchmarks** | [Launch / B5](https://www.supercompress.dev/benchmarks) | [Legacy section](https://www.supercompress.dev/benchmarks#legacy-compiler) |
  62. ### Two products
  63. | | **Coding-agent plugin** | **API / Python** |
  64. |---|---|---|
  65. | **For** | Cursor, Claude Code, Codex, and 40+ agent harnesses | Apps, RAG, agents, backends |
  66. | **Install** | `npm i -g supercompress-proxy && npx supercompress setup` | `pip install supercompress` |
  67. | **What you get** | MCP `compress_context` on big dumps | Compress before every model call |
  68. | **Login** | Keep your normal agent login | API key from the [dashboard](https://www.supercompress.dev/dashboard) |
  69. Docs: [coding agents (first 5 minutes checklist)](https://docs.supercompress.dev/coding-agents#quickstart) · [API quickstart](https://docs.supercompress.dev/quickstart)
  70. ### Repo map
  71. | Path | What |
  72. |------|------|
  73. | `packages/proxy` | Coding-agent plugin (npm) |
  74. | `api/` | Hosted API + billing |
  75. | `web/` | Site + docs HTML |
  76. | `supercompress/` | Python package |
  77. | `docs/REPO_LAYOUT.md` | What belongs in OSS vs private |
  78. Private marketing, outreach, and model training stay **out** of this repo (see `.gitignore` + `docs/REPO_LAYOUT.md`).
  79. ---
  80. ## Benchmarks & stats
  81. We measure **whether required evidence survives** (containment), not downstream LLM completion.
  82. ### Neural v2 launch (hosted ~400M) — B5 coding-agent suite
  83. | Metric | Result |
  84. |---|---:|
  85. | **Mean context cut** | **64.1%** |
  86. | **Evidence passes** | **24 / 24** |
  87. | **Tokens** | **16,647 → 5,148** |
  88. | **Max cut @ ≥99% retention** | **96.6%** |
  89. | **B5 latency p50** | **~5.8 s** (hosted Neural Keep on Fly CPU) |
  90. | **Public cases (full suite)** | **390** |
  91. | **Downstream LLM eval** | **Not yet run** |
  92. Aggregate mean cut across all 390 cases is ~3.6% — the engine often refuses to over-cut dense needle/QA slices. The 64.1% figure is the coding-agent suite where dumps are noisy.
  93. Raw JSON: [launch-benchmark.json](https://www.supercompress.dev/assets/data/launch-benchmark.json) · writeup: [benchmarks](https://www.supercompress.dev/benchmarks)
  94. ### Legacy compiler (separate product)
  95. CPU / millisecond-class local path. Older held-out compiler numbers (≈58–66% cut, 99.4% gold containment) live under [Legacy/compiler on /benchmarks](https://www.supercompress.dev/benchmarks#legacy-compiler). **Do not mix with Neural v2.**
  96. ---
  97. ## Try it
  98. ### Coding agents (recommended)
  99. ```bash
  100. npm install -g supercompress-proxy
  101. npx supercompress setup
  102. ```
  103. Links your account, detects agents, installs MCP. Docs: [coding agents](https://docs.supercompress.dev/coding-agents)
  104. ### Python / HTTP
  105. ```bash
  106. pip install supercompress
  107. export SUPERCOMPRESS_API_KEY=sc_live_YOUR_KEY
  108. ```
  109. ```python
  110. from supercompress.client import SuperCompress
  111. sc = SuperCompress()
  112. result = sc.compress(
  113. context=long_context,
  114. query="What failed and how do we fix it?",
  115. )
  116. print(f"{result.original_tokens} → {result.kept_tokens} tokens")
  117. print(result.compressed_text)
  118. ```
  119. ```bash
  120. curl -X POST https://www.supercompress.dev/api/v1/compress \
  121. -H "X-API-Key: $SUPERCOMPRESS_API_KEY" \
  122. -H "Content-Type: application/json" \
  123. -d '{"context":"...","query":"What failed?"}'
  124. ```
  125. Or paste a dump into the **[Arena](https://www.supercompress.dev/arena)** — no integration required.
  126. ---
  127. ## How it compares
  128. | | Truncate | Summarize | **SuperCompress** |
  129. |---|:---:|:---:|:---:|
  130. | Cuts tokens | ✓ | ✓ | ✓ |
  131. | Uses the query | ✗ | weak | ✓ |
  132. | Keeps original evidence | sometimes | ✗ | ✓ |
  133. | Auditable kept lines | partial | ✗ | ✓ |
  134. More: [vs truncation](https://www.supercompress.dev/supercompress-vs-truncation) · [vs summarization](https://www.supercompress.dev/supercompress-vs-summarization) · [vs alternatives](https://www.supercompress.dev/supercompress-vs-alternatives)
  135. ---
  136. <p align="center">
  137. <a href="https://www.supercompress.dev/arena"><img src="https://img.shields.io/badge/Try_the_Arena-2563EB?style=for-the-badge" alt="Arena" /></a>
  138. &nbsp;
  139. <a href="https://www.supercompress.dev/dashboard"><img src="https://img.shields.io/badge/Get_an_API_key-111827?style=for-the-badge" alt="Dashboard" /></a>
  140. &nbsp;
  141. <a href="https://docs.supercompress.dev/coding-agents"><img src="https://img.shields.io/badge/Install_for_agents-059669?style=for-the-badge" alt="Agents" /></a>
  142. </p>
  143. <p align="center">
  144. <sub>
  145. <a href="https://www.supercompress.dev">supercompress.dev</a> ·
  146. <a href="./LICENSE">MIT License</a> ·
  147. <a href="https://www.supercompress.dev/dashboard?signup=1">Sponsor</a> ·
  148. built by <a href="https://github.com/arjunkshah12345-hash">Arjun Shah</a>
  149. </sub>
  150. </p>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.