Laporan audit skill

sparse-autoencoder-training Laporan audit.

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.

Eksperimental · TinjauPerlu ditinjauDihasilkan 13 Okt 2026Audit metadata heuristik

Audit

Kepercayaan

Kualitas

Keamanan

100

Maintain

Pasang

Trust Score OpenAgentSkill

Shortlist kuat

Trust Score OpenAgentSkill

The Trust Score helps an agent decide whether a skill is safe enough to shortlist before installation.

Adopsi GitHub

Lulus

100

13K star GitHub

Aktivitas star/fork

Lulus

13K star dan 948 fork; aktivitas issue tidak tersedia dalam metadata saat ini

Pemeliharaan terbaru

Lulus

100

2 hari sejak push

Kejelasan lisensi

Lulus

MIT

Kelengkapan README/SKILL.md

Lulus

Metadata memuat konteks penggunaan dan alur kerja yang cukup

Risiko dependensi/runtime

Gagal

command execution surface, credential or environment access

Ketersediaan pemasangan

Lulus

npx skills add Orchestra-Research/AI-Research-SKILLs --skill sparse-autoencoder-training

Keamanan perintah pemasangan

Lulus

Jalur pemasangan paket atau runtime standar

Cakupan izin

Gagal

secrets or environment access, shell or command execution

Bukti repositori

Lulus

https://github.com/Orchestra-Research/AI-Research-SKILLs/tree/main/04-mechanistic-interpretability/saelens

Status peninjauan

Peringatan

50.9

Persetujuan tinjauan AI belum ada

Hasil terbukti Agent

Info

Belum ada data hasil Agent

Pemeriksaan

Tinjauan pemasangan dan adopsi

8 Lulus · 11 Perlu ditinjau

Jalur pemasangan

Lulus

npx skills add Orchestra-Research/AI-Research-SKILLs --skill sparse-autoencoder-training

Repositori

Lulus

https://github.com/Orchestra-Research/AI-Research-SKILLs/tree/main/04-mechanistic-interpretability/saelens

Lisensi

Lulus

MIT

Pemeliharaan

100

Lulus

2 hari sejak push

Tinjauan AI

Periksa

Review approval is missing

Kelengkapan README/SKILL.md

Lulus

Usable description available

Risiko dependensi

Perbaiki

command execution surface, credential or environment access

Keamanan perintah pemasangan

Lulus

Jalur pemasangan paket atau runtime standar

Cakupan izin

Perbaiki

secrets or environment access, shell or command execution

Aktivitas star/fork

Lulus

13K star dan 948 fork; aktivitas issue tidak tersedia dalam metadata saat ini

Adopsi

Lulus

13K star GitHub

Peringatan

Dependency or permission surface needs review
Permission surface may require sandboxing
Persetujuan tinjauan AI belum ada
Quality score needs review
Permission surface needs review: secrets or environment access, shell or command execution
Dependency/runtime risk: command execution surface, credential or environment access
Permission surface: secrets or environment access, shell or command execution
Review status: AI review approval is missing

Metode

This report combines public metadata, AI review output, repository freshness, install readiness, OpenAgentSkill events, quality scoring, trust checks, and the agent safety gate. It is not a full source-code security review.

Jalur pemasangan

Tinjau laporan sebelum memasang pada Agent produksi.

$ npx skills add Orchestra-Research/AI-Research-SKILLs --skill sparse-autoencoder-training

Metadata yang dapat dibaca Agent

Paket keputusan mesin

JSON tersemat menyediakan tugas dan Agent yang sesuai, perintah pemasangan, skor kepercayaan, gerbang keamanan, alternatif, serta kondisi jangan gunakan untuk ditinjau Agent.

Kebijakan pemasangan: Tinjau
Agent terbaik: Codex + Claude Code
Alternatif: 3

Buka JSON Agent

Sinyal

Skor audit: 84/100
Quality: 87/100
Trust: 78/100
Install safety: 92/100
Cakupan izin: 22/100
Pemeliharaan: 2 hari sejak push
Event: 2 event
Star GitHub: 13K
Push terakhir: 11 Okt 2026

Keamanan Agent v2

44/100 · Hindari pemasangan otomatis

EksperimentalTinjau

Sparse or mixed signals. Useful for discovery, but not for autonomous installation.

Test manually in an isolated workspace and compare against safer alternatives.

Metadata combines secrets access with shell or command execution
Petunjuk izin berisiko tinggi: Shell or command execution, Secrets or environment access
skor keamanan Agent 44/100

Eksekusi shell atau perintah

Tinggi

Metadata skill merujuk terminal, CLI, shell, subprocess, atau alur kerja eksekusi perintah.

Akses jaringan

Sedang

Skill kemungkinan mengambil halaman jarak jauh, API, repositori, atau layanan eksternal.

Akses sistem file

Sedang

Skill dapat membaca atau menulis file proyek, dokumen, artefak yang dihasilkan, atau status workspace lokal.

Secrets or environment access

Tinggi

Skill metadata references credentials, tokens, environment variables, or secret-bearing workflows.

Petunjuk izin berisiko tinggi: Shell or command execution, Secrets or environment access
Dependency or permission surface needs review

Kembali ke skill Alternatif Badge audit SVG

Bandingkan opsi sekitar

Skill terkait untuk diaudit berikutnya

weather

Current weather and forecasts with web_fetch, falling back to wttr.in curl for locations, rain, temperature, travel planning.

391K Star · Laporan audit

mcporter

List, configure, authenticate, call, and inspect MCP servers/tools with mcporter over HTTP or stdio.

391K Star · Laporan audit

release-openspec

Use this skill when releasing OpenSpec: audit merged work and changeset coverage, decide whether a catch-up changeset PR is needed, prepare or resume the Changesets Version Packages PR, cut a beta or stable release, verify publishing, and polish GitHub release notes. Also use when asked whether an open release PR is complete, what the next release step is, or to continue a release paused for human approval.

71K Star · Laporan audit