📦

agent-browser

Name: agent-browser
Author: inference-sh-0

Révision du contenu r1 Critique ⚙️ Commandes externes🌐 Accès réseau📁 Accès au système de fichiers

Automatiser les tâches de navigateur avec des agents IA

Les tâches web nécessitent souvent une navigation répétée, la saisie de formulaires et des vérifications visuelles. Cette skill fournit à Claude, Codex et Claude Code un workflow de navigateur structuré via inference.sh.

Prend en charge: Claude Codex Code(CC)

⚠️ 38 Médiocre

Maintain this Skill? Claim it ⚠️ 38 Médiocre

N’installez pas automatiquement ce Skill.

La politique de référence exige un examen par un opérateur avant toute action d’installation.

Installer avec mon Agent

Copiez cette demande dans votre Agent. Elle contient la page Skill de référence et le manifeste.

Demande d’agent

Review the Skillstore skill "agent-browser" from https://skillstore.io/skills/inference-sh-0-agent-browser.md and its manifest at https://skillstore.io/api/skills/inference-sh-0-agent-browser/manifest. Verify the artifact. Do not auto-install. Inspect the skill and report your findings, then wait for an operator or manual installation decision.

Votre Agent doit toujours présenter son plan et demander toute confirmation exigée par la politique de sécurité.

Ressources lisibles par les agents

Utilisez ces liens lorsqu’un Agent IA, un crawler ou un script a besoin d’un contexte propre au lieu de lire toute la page.

Détail Markdown GET /skills/inference-sh-0-agent-browser.md manifest signé GET /api/skills/inference-sh-0-agent-browser/manifest lockfile signé GET /api/skills/inference-sh-0-agent-browser/lockfile

Tester

Utilisation de « agent-browser ». Inspecter une page de tarification et capturer les appels à l’action visibles.

Résultat attendu:

La page a été ouverte et le sélecteur de forfait principal, le bouton de bascule de facturation et le bouton de contact commercial ont été trouvés.
Une capture d’écran de la fenêtre d’affichage a été capturée et il a été noté que le forfait entreprise nécessite un flux de contact séparé.
Aucun formulaire n’a été soumis et aucun contenu authentifié n’a été consulté.

Utilisation de « agent-browser ». Remplir un formulaire de contact de test avec des exemples de données approuvés.

Résultat attendu:

Les contrôles de nom, d’e-mail, de message et d’envoi ont été identifiés à partir du snapshot actuel.
Les valeurs de test approuvées ont été saisies et le formulaire a été soumis.
Le message de confirmation est apparu après la soumission, et la session a été fermée.

Utilisation de « agent-browser ». Enregistrer un court workflow pour une revue de documentation.

Résultat attendu:

L’enregistrement a démarré avant le workflow approuvé et s’est arrêté à la fermeture de la session.
Des captures d’écran clés ont été capturées pendant les étapes de navigation.
Une étape a été signalée pour examen manuel, car le bouton attendu n’était pas visible.

Audit de sécurité

Critique

v4 • 05/07/2026 Ouvrir le rapport versionné

The skill provides useful browser automation, but it relies on shell-driven infsh workflows, remote browser capture, proxy support, file upload, and page JavaScript execution. The critical issue is the pipe-to-shell installer; additional high-risk content includes cookie extraction guidance and proxy rotation for rate-limit avoidance. No prompt injection attempt was found in the reviewed files.

Fichiers analysés

2,105

Lignes analysées

Éléments à examiner

Faux positifs ignorés

Préoccupations de sécurité confirmées (6)

Critique

SKILL.md:17

Pipe to shell pattern

curl -fsSL https://cli.inference.sh | sh && infsh login

The quick start pipes a remote installer from cli.inference.sh directly into sh. This can execute unverified network content on the user machine.

Élevé

SKILL.md:3

Screen capture upload

description: "Browser automation for AI agents via inference.sh. Navigate web pages, interact with e

The skill description advertises screenshots and video recording for browser sessions. Captures can include credentials or private page content and may be returned through the external service.

Élevé

references/authentication.md:209-224

Cookie Extraction Guidance

The authentication reference shows extracting browser cookies and printing them after an authenticated session. This can expose session tokens or private account data outside the browser.

The section title and examples directly describe cookie extraction from an authenticated browser session. The risk is clear even though the code is presented as documentation.

Élevé

references/proxy-support.md:85-129

Rate Limit Evasion via Proxy Rotation

The proxy guide includes a rate-limit avoidance workflow that rotates proxies for scraping. This supports evasion of site controls and can enable abusive automation.

The heading and example explicitly connect proxy rotation with rate-limit avoidance. This is strong intent-level evidence beyond generic proxy support.

Élevé

SKILL.md:154-163

Local File Upload Capability

The skill supports uploading local file paths into browser file inputs. Without explicit user review, an agent could disclose local files to an external website.

The documented upload action accepts file paths and sends them through a browser form. The capability is legitimate but has a clear data disclosure risk.

Moyen

SKILL.md:178-185 references/commands.md:162-176

Arbitrary Page JavaScript Execution

The skill exposes an execute function for running custom JavaScript in the current page. This can extract page data or alter page state during authenticated sessions.

The function is documented as accepting custom JavaScript code. The browser context makes the risk concrete, but it is also an intended automation feature.

Éléments d’évaluation des capacités (84)

Il s’agit de capacités locales réelles qui peuvent être attendues pour cette compétence ; elles nécessitent donc un examen, mais ne sont pas comptabilisées comme comportement malveillant confirmé.

Élevé

references/authentication.md:215-217

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION --input '{

The authentication guide shows extracting and printing cookie-related browser data from an authenticated session. That can expose session secrets outside the browser context.

Élevé

references/authentication.md:218

Shell command substitution

COOKIES=$(echo $RESULT | jq -r '.result')

The authentication guide shows extracting and printing cookie-related browser data from an authenticated session. That can expose session secrets outside the browser context.

Élevé

references/authentication.md:222-223

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION --input '{

The authentication guide shows extracting and printing cookie-related browser data from an authenticated session. That can expose session secrets outside the browser context.

Élevé

references/proxy-support.md:152

SOCKS proxy

"proxy_url": "socks5://privacy-proxy.example.com:1080"

The proxy documentation explicitly supports SOCKS proxy routing. This can hide traffic origin and is paired with scraping and anonymity workflows in the same reference.

Élevé

references/proxy-support.md:165

SOCKS proxy

### SOCKS5 Proxy

The proxy documentation explicitly supports SOCKS proxy routing. This can hide traffic origin and is paired with scraping and anonymity workflows in the same reference.

Élevé

references/proxy-support.md:168

SOCKS proxy

{"proxy_url": "socks5://proxy.example.com:1080"}

The proxy documentation explicitly supports SOCKS proxy routing. This can hide traffic origin and is paired with scraping and anonymity workflows in the same reference.

Élevé

SKILL.md:17

Hardcoded URL

curl -fsSL https://cli.inference.sh | sh && infsh login

The URL is used to download the CLI installer that is piped into a shell. This is a real network trust and code execution risk.

Moyen

references/authentication.md:24-26

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:51

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:52

Shell command substitution

URL=$(echo $RESULT | jq -r '.url')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:70-72

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:80

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:97

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:115

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:116

Shell command substitution

URL=$(echo $RESULT | jq -r '.url')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:128

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:129

Shell command substitution

ELEMENTS=$(echo $RESULT | jq -r '.elements_text')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:133

Shell command substitution

TOTP_CODE=$(oathtool --totp -b "$TOTP_SECRET")

The example invokes oathtool with a TOTP secret during automated login. This is legitimate but processes authentication secrets through shell commands.

Moyen

references/authentication.md:153-156

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:179-181

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:202

Shell command substitution

SESSION=$(login)

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:243

Shell command substitution

export PASSWORD=$(cat /path/to/secure/password)

The example reads passwords from local files or a secrets manager into shell environment variables. This is legitimate guidance but still handles sensitive credentials in shell state.

Moyen

references/authentication.md:246

Shell command substitution

export PASSWORD=$(vault read -field=password secret/app)

The example reads passwords from local files or a secrets manager into shell environment variables. This is legitimate guidance but still handles sensitive credentials in shell state.

Moyen

references/authentication.md:285

Shell command substitution

URL=$(echo $RESULT | jq -r '.url')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/authentication.md:292

Shell command substitution

ELEMENTS=$(echo $RESULT | jq -r '.elements_text')

The finding is a shell command substitution in skill-provided instructions. Legitimate use is possible, but the pattern executes commands and is kept fail-closed.

Moyen

references/proxy-support.md:22-25

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:35-40

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:65-68

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:76

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION --input '{}')

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:113-116

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:119-121

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:137-142

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:150-153

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:186-189

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:192-194

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:195

Shell command substitution

echo "IP via proxy: $(echo $RESULT | jq -r '.result')"

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:267-270

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/proxy-support.md:273

Shell command substitution

SESSION_ID=$(echo $SESSION | jq -r '.session_id')

The proxy reference includes executable shell examples that run infsh through proxy settings. This is intended functionality, but it enables external network automation through a shell workflow.

Moyen

references/snapshot-refs.md:124-126

Shell command substitution

RESULT=$(infsh app run agent-browser --function interact --session $SESSION --input '{

The documentation shows command substitution around infsh interaction examples. It is intended usage but still executes an external browser automation CLI.

Moyen

references/snapshot-refs.md:136-138

Shell command substitution

RESULT=$(infsh app run agent-browser --function interact --session $SESSION --input '{

The documentation shows command substitution around infsh interaction examples. It is intended usage but still executes an external browser automation CLI.

Moyen

references/video-recording.md:23-26

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:38

Shell command substitution

RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:39

Shell command substitution

VIDEO=$(echo $RESULT | jq -r '.video')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:48-52

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:87-90

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:93-95

Shell command substitution

RESULT=$(infsh app run agent-browser --function interact --session $SESSION --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:97

Shell command substitution

SUCCESS=$(echo $RESULT | jq -r '.success')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:100

Shell command substitution

echo "Message: $(echo $RESULT | jq -r '.message')"

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:103

Shell command substitution

CLOSE_RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:104

Shell command substitution

echo "Debug video: $(echo $CLOSE_RESULT | jq -r '.video')"

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:119-124

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:156

Shell command substitution

RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:157

Shell command substitution

echo "Documentation video: $(echo $RESULT | jq -r '.video')"

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:168-171

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:178

Shell command substitution

CLOSE_RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:179

Shell command substitution

VIDEO=$(echo $CLOSE_RESULT | jq -r '.video')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:195

Shell command substitution

TASK_ID=$(date +%Y%m%d-%H%M%S)

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:197-200

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:206

Shell command substitution

RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

references/video-recording.md:207

Shell command substitution

VIDEO=$(echo $RESULT | jq -r '.video')

The video reference includes executable shell examples that start browser sessions and collect recording output. These workflows can capture sensitive page activity if run on private sessions.

Moyen

SKILL.md:34-36

Shell command substitution

RESULT=$(infsh app run agent-browser --function open --session new --input '{

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:37

Shell command substitution

SESSION_ID=$(echo $RESULT | jq -r '.session_id')

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:114-118

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:213-215

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:231-233

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:246-249

Shell command substitution

SESSION=$(infsh app run agent-browser --function open --session new --input '{

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

SKILL.md:257

Shell command substitution

RESULT=$(infsh app run agent-browser --function close --session $SESSION --input '{}')

The main skill instructions include executable Bash examples using command substitution with infsh. This is intended usage but still asks agents or users to run external commands.

Moyen

templates/authenticated-session.sh:40-42

Shell command substitution

RESULT=$(infsh app run agent-browser --function open --session new --input '{

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/authenticated-session.sh:43

Shell command substitution

SESSION_ID=$(echo $RESULT | jq -r '.session_id')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/authenticated-session.sh:1

Unix shell invocation

#!/bin/bash

This is an executable Bash template shipped with the skill. The script invokes the infsh browser automation CLI and can act on user-provided web targets.

Moyen

templates/capture-workflow.sh:28

Shell command substitution

CLOSE_RESULT=$(infsh app run agent-browser --function close --session $SESSION_ID --input '{}' 2>/de

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:31

Shell command substitution

VIDEO=$(echo $CLOSE_RESULT | jq -r '.video // empty')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:54-59

Shell command substitution

RESULT=$(infsh app run agent-browser --function open --session new --input '{

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:60

Shell command substitution

SESSION_ID=$(echo $RESULT | jq -r '.session_id')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:63

Shell command substitution

URL=$(echo $RESULT | jq -r '.url')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:64

Shell command substitution

TITLE=$(echo $RESULT | jq -r '.title')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:86

Shell command substitution

RESULT=$(infsh app run agent-browser --function snapshot --session $SESSION_ID --input '{}')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:105-107

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION_ID --input '{

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:114-115

Shell command substitution

RESULT=$(infsh app run agent-browser --function execute --session $SESSION_ID --input '{

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/capture-workflow.sh:1

Unix shell invocation

#!/bin/bash

This is an executable Bash template shipped with the skill. The script invokes the infsh browser automation CLI and can act on user-provided web targets.

Moyen

templates/form-automation.sh:30-32

Shell command substitution

RESULT=$(infsh app run agent-browser --function open --session new --input '{

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/form-automation.sh:33

Shell command substitution

SESSION_ID=$(echo $RESULT | jq -r '.session_id')

This is active template code that runs infsh or parses its output through shell command substitution. The template can browse user-supplied URLs through a network-backed CLI.

Moyen

templates/form-automation.sh:1

Unix shell invocation

#!/bin/bash

This is an executable Bash template shipped with the skill. The script invokes the infsh browser automation CLI and can act on user-provided web targets.

Faible

SKILL.md:11

Hardcoded URL

![Agentic Browser](https://cloud.inference.sh/app/files/u/4mg21r6ta37mpaz6ktzwtt8krr/01kgjw8atdxgkrs

The Markdown loads a remote image from cloud.inference.sh. This is low severity, but it can create an external request when rendered.

Facteurs de risque

⚙️ Commandes externes (162)

🌐 Accès réseau (54)

📁 Accès au système de fichiers (15)

references/authentication.md:5 references/authentication.md:263 references/commands.md:3 references/commands.md:59 references/proxy-support.md:5 references/proxy-support.md:272 references/snapshot-refs.md:5 references/video-recording.md:5 references/video-recording.md:239 templates/authenticated-session.sh:30 templates/capture-workflow.sh:28 templates/capture-workflow.sh:78 templates/capture-workflow.sh:81 templates/capture-workflow.sh:149 templates/form-automation.sh:23

Motifs détectés

Screen capture uploadPipe to shell pattern

Audité par: codex Voir l’historique des audits →

Partager et citer ce rapport

Partagez le rapport d’évaluation versionné, le badge neutre, la carte intégrable et les citations. Skillstore présente les preuves sans décider si ce Skill est sûr.

Ouvrir le rapport versionné

Copier le lien du rapport

https://skillstore.io/skills/inference-sh-0-agent-browser/audits/4?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Badge Markdown

[![Skillstore security assessment](https://skillstore.io/badges/skills/inference-sh-0-agent-browser/security.svg)](https://skillstore.io/skills/inference-sh-0-agent-browser?utm_source=security_passport_badge)

Badge HTML

<a href="https://skillstore.io/skills/inference-sh-0-agent-browser?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/inference-sh-0-agent-browser/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Carte intégrable

<iframe src="https://skillstore.io/embed/skills/inference-sh-0-agent-browser.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Citations académiques (APA · BibTeX · CFF)

Citation APA

inference-sh-0. (2026). agent-browser security audit report (audit version 4) [Author version unspecified]. Skillstore. https://skillstore.io/skills/inference-sh-0-agent-browser/audits/4

Citation BibTeX

@techreport{inference-sh-0-inference-sh-0-agent-browser-2026,
  author = {inference-sh-0},
  title = {agent-browser security audit report (audit version 4)},
  institution = {Skillstore},
  year = {2026},
  number = {4},
  url = {https://skillstore.io/skills/inference-sh-0-agent-browser/audits/4},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "agent-browser security audit report (audit version 4)"
version: "unspecified"
type: report
authors:
  - name: "inference-sh-0"
date-released: "2026-07-05"
url: "https://skillstore.io/skills/inference-sh-0-agent-browser/audits/4"
identifiers:
  - type: other
    value: "skillstore:inference-sh-0-agent-browser:audit:4"
    description: "Skillstore immutable audit report identifier"

Comparer les variantes

19 variantes installables

Chaque auteur reste un skill installable distinct. La variante recommandée est classée selon les preuves Skillstore.

Pourquoi cette variante est première

Meilleur score Skillstore

Auteur / source Score Skillstore Niveau de confiance des preuves Utilisation Skillstore Mis à jour

supercent-io Recommandé

supercent-io-agent-browser

Score Skillstore 78

Niveau de confiance des preuves Élevé

Utilisation Skillstore 52

Mis à jour

2026-09-09

101-skills

101-skills-agent-browser

Score Skillstore 74

Niveau de confiance des preuves Moyen

Utilisation Skillstore 10

Mis à jour

2026-10-12

vercel-labs

vercel-labs-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Élevé

Utilisation Skillstore 91

Mis à jour

2026-09-09

qu-skills

qu-skills-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 46

Mis à jour

2026-09-09

tul-sh

tul-sh-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 44

Mis à jour

2026-09-09

am-will

am-will-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 43

Mis à jour

2026-10-23

inference-sh-8

inference-sh-8-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 32

Mis à jour

2026-09-09

skillssh

skillssh-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 23

Mis à jour

2026-09-09

toolshell

toolshell-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 22

Mis à jour

2026-09-09

inference-sh-9

inference-sh-9-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 21

Mis à jour

2026-09-09

infsh-skills

infsh-skills-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 18

Mis à jour

2026-09-09

inferen-sh

inferen-sh-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 17

Mis à jour

2026-09-09

inference-sh-skills

inference-sh-skills-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 16

Mis à jour

2026-09-09

inferencesh

inferencesh-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 16

Mis à jour

2026-09-09

inference-sh-0 Actuelle

inference-sh-0-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 16

Mis à jour

2026-09-09

inference-shell

inference-shell-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 15

Mis à jour

2026-09-09

inf-sh

inf-sh-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 14

Mis à jour

2026-09-09

inference-skills

inference-skills-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 14

Mis à jour

2026-09-09

halt-catch-fire

halt-catch-fire-agent-browser

Score Skillstore 38

Niveau de confiance des preuves Moyen

Utilisation Skillstore 8

Mis à jour

2026-09-09

Score Skillstore

Pourquoi ce score Niveau de confiance des preuves: Moyen

Architecture

Maintenabilité

Contenu

Communauté

Conformité aux spécifications

Ce que vous pouvez créer

Vérifier les workflows web

Exécuter des actions de navigateur répétables, capturer des captures d’écran et vérifier les résultats visibles après chaque étape.

Collecter des preuves de page

Ouvrir des pages publiques, extraire le texte et les liens visibles, et enregistrer des captures d’écran pour examen.

Automatiser les tâches d’administration

Naviguer dans des outils authentifiés, remplir des formulaires et enregistrer des workflows sélectionnés après approbation explicite.

Essayez ces prompts

Ouvrir et résumer une page

Ouvre cette URL avec agent-browser, capture l’état visible de la page et résume les principaux éléments interactifs : [URL]

Remplir un formulaire simple

Utilise agent-browser pour inspecter ce formulaire, identifier les refs @e nécessaires, remplir les valeurs approuvées, l’envoyer et rapporter le résultat : [URL]

Capturer des preuves pour un workflow

Exécute ce workflow de navigateur approuvé, prends des captures d’écran après les étapes clés et explique toute action échouée sans réessayer les étapes destructrices : [WORKFLOW]

Auditer une session authentifiée

Après ma confirmation de la cible et des identifiants, utilise agent-browser pour effectuer les vérifications authentifiées approuvées, évite d’enregistrer les secrets et ferme la session une fois terminé.

Bonnes pratiques

Confirmer l’URL cible, la sensibilité des données et les actions autorisées avant d’ouvrir une session de navigateur.
Effectuer un nouveau snapshot après une navigation ou des mises à jour dynamiques de la page afin que chaque action utilise les refs @e actuelles.
Désactiver l’enregistrement vidéo lors de la manipulation d’identifiants, de données privées ou de contenu réglementé.

Éviter

Ne pas exécuter l’installateur distant via un pipe to shell sans vérification.
Ne pas utiliser la rotation de proxy pour contourner les limites de débit, les contrôles d’accès ou les politiques des sites.
Ne pas extraire de cookies, téléverser des fichiers locaux ni exécuter JavaScript dans la page sans approbation explicite.

Foire aux questions

Cette skill nécessite-t-elle un autre service ?

Oui. Elle nécessite la CLI inference.sh et un compte infsh authentifié.

Peut-elle interagir avec des pages connectées ?

Oui, mais les identifiants et les données de session nécessitent l’approbation explicite de l’utilisateur et une manipulation prudente.

Que sont les refs @e ?

Ce sont des références d’éléments compactes renvoyées par les snapshots afin que les agents puissent cliquer, remplir et sélectionner les contrôles de page.

Peut-elle enregistrer une vidéo ?

Oui. L’enregistrement est facultatif et doit rester désactivé pour les sessions sensibles.

Prend-elle en charge les proxy ?

Oui. La prise en charge des proxy est documentée pour les tests et l’accès en entreprise, mais une mauvaise utilisation peut enfreindre les politiques des sites.

Est-elle sûre pour des sites web arbitraires ?

Faites preuve de prudence. L’automatisation de navigateur peut déclencher des actions réelles, exposer des données de page ou téléverser des fichiers en cas de mauvaise utilisation.