Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    PHP2Shell Audit | CVE-2026-109618|2026-109619|2026-115560

    2

    PHP projects running outdated versions face significant risk from recent CVEs that allow remote shell access.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    PHP2Shell Audit | CVE-2026-109618|2026-109619|2026-115560

    PHP2Shell Audit | CVE-2026-109618|2026-109619|2026-115560

    Example session with this skill installed

    Check whether https://example.com is running a PHP version affected by php2shell (CVE-2026-109618 and CVE-2026-109619). Run the passive check first and tell me the PHP version, whether a CDN may hide it, and what I should do next. If an active test is needed, ask me for confirmation before sending anything.

    • Read your context and instructions
    • Compiled the php2shell audit
    • Generated the document

    Example output — passive check against a public site.

    Summary
    PHP 8.4.19 was detected on https://example.com. That version is below the fixed release for the 8.4 branch (8.4.27), so the site may be vulnerable to php2shell (CVE-2026-109618 / CVE-2026-109619). The check was passive: headers, a limited body read, and an exposed phpinfo. No POST request or payload was sent.

    Findings

    ItemResult
    Targethttps://example.com
    ReachableYes (HTTP 200)
    PHP version8.4.19
    Fixed version for 8.48.4.27 or later
    Version sourceX-Powered-By header
    SAPIFPM/FastCGI
    CDN / edgeNone detected
    Exposed phpinfoNo
    Evidence levelVersion only
    ConfidenceMedium

    What this means
    This is not a confirmed exploit. It is a version-based warning. A CDN, proxy, or a spoofed X-Powered-By header could make the version unreliable. The active crash test only runs against localhost, .local, .test, or private IPs, and only after explicit confirmation.

    Next steps

    1. Ask the hosting provider or platform to update PHP to 8.4.27 or later.
    2. Restart PHP-FPM after the update.
    3. Re-run the passive check to confirm the reported version.
    4. If you can reproduce the site in staging, run the active test there with authorization.
    5. If the site was exposed before the update, run the read-only triage to look for signs of exploitation.

    Report
    RESULT=POSSIBLY_VULNERABLE
    CAUSE=VERSION_OUTDATED
    HTTP=200
    VERSION=8.4.19 SAPI=FPM/FastCGI EDGE=none
    EVIDENCE=PHP 8.4.19 is below the fixed release 8.4.27 for the 8.4 branch (X-Powered-By).
    CONFIDENCE=medium
    NEXT=Update PHP to 8.4.27 or later, restart PHP-FPM, and confirm the version. Use local/staging for the active test.
    REPORT=/home/user/.local/state/php2shell-scans/example.com-20261101T210700Z.report.json
    REPORT_MD=/home/user/.local/state/php2shell-scans/example.com-20261101T210700Z.report.md
    REPORT_HTML=/home/user/.local/state/php2shell-scans/example.com-20261101T210700Z.report.html

    report.html

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    PHP projects running outdated versions face significant risk from recent CVEs that allow remote shell access. This skill provides a structured auditing workflow to identify vulnerabilities and signs of exploitation without manual scanner configuration.

    What it does

    • Vulnerability scanning audits targets for CVE-2026-109618, CVE-2026-109619, and CVE-2026-115560.
    • Passive reconnaissance performs version checks and header analysis on public URLs without sending intrusive payloads.
    • Active testing executes high-traffic probes against local or staging environments to confirm worker crashes.
    • Incident triage inspects local logs and file systems for indicators of prior exploitation.
    • Configuration audit analyzes PHP session settings to identify insecure serialization handlers.

    How it works

    1. Target Identification starts by asking for a public URL or a local project path and its corresponding local URL.
    2. Environment Preflight verifies system dependencies like uv, git, and curl before installing the engine.
    3. Engine Deployment clones the audit repository and downloads pinned versions of the Assetnote scanner and triage scripts.
    4. Execution runs passive checks for public targets or active, confirmed tests for local targets.
    5. Reporting generates a normalized summary including the result, evidence, and specific update paths.

    Frameworks & tools

    The skill relies on bash for orchestration, uv for Python environment management, and git for versioned tool retrieval. It audits PHP environments including mod_php and PHP-FPM setups.

    Why this beats prompting it yourself

    Manually running security scanners requires managing complex dependencies and pinned commits to ensure integrity. This skill automates the environment setup, enforces strict safety boundaries between public and local targets, and normalizes output into actionable remediation steps.

    Use cases

    • Verify if a production site is exposing a vulnerable PHP version string.
    • Perform a stress-test audit on a local staging server before deployment.
    • Search for post-exploitation traces in system logs after a suspected breach.
    • Audit php.ini session configurations for insecure serialization logic.

    Known limitations

    Requires bash and uv installed on the host. Active tests can temporarily take down PHP services and require explicit user confirmation. Does not replace professional security audits.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.

    Requires bash, git, curl, and uv. Active testing may cause 502/503 errors. Only supports passive checks for public URLs. Windows users need Git Bash or WSL2.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 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 15 days ago

    Software
    BashGitcurluv (Python tool runner)Internet access
    Optional
    Python 3PHP

    On Windows use Git Bash or WSL2. uv can provide Python. Internet access must reach GitHub, gist.githubusercontent.com and PyPI.

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Listed15 days ago

    What's inside

    Frequently Asked Questions

    php2shell is a PHP vulnerability that can lead to remote code execution in affected configurations. This skill covers CVE-2026-109618 and CVE-2026-109619, plus the session-configuration variant CVE-2026-115560.

    It asks for a public URL or a local PHP project, checks the required tools, installs the pinned engine, and runs the appropriate flow. Public targets get passive checks only. Local targets get an active crash test after explicit confirmation. It also offers a read-only triage for signs of prior exploitation.

    Linux, macOS, WSL2, or Windows with Git Bash. Required: bash, git, curl, uv, and sha256sum or shasum. Python 3 is recommended (uv can provide it). PHP is optional: remote passive checks do not need it, but local project audits work better with it. Internet access to GitHub, gist.githubusercontent.com, and PyPI is needed during setup.

    No. It only needs an agent that can run shell commands and read files. A structured question tool is optional. No MCP server, plugin, or desktop app is required.

    For public URLs it only performs passive checks: headers, a limited body read, and an exposed phpinfo, with a maximum of 5 GET requests. It never sends POST or payloads there. The active crash test must not be run against production: use local, .local, .test, or private IPs, and only with written authorization.

    It can. It sends up to about 29 MB and may take PHP-FPM or mod_php down for a few seconds. That is why it only runs on local or staging targets after explicit confirmation, and why the skill offers a dry-run first.

    No. It sends controlled crash probes to confirm whether the target is affected. It does not provide a shell, dump data, or display exploit payloads. It reports evidence and remediation steps.

    8.2.35, 8.3.36, 8.4.27, and 8.5.12. Branches 5.x, 7.x, 8.0, and 8.1 have no upstream patch and are treated as vulnerable.

    It can run a read-only triage that looks for signs in logs, sessions, webroot files, processes, cron, SSH keys, and network connections. It does not prove a compromise and does not clean anything. Updating PHP does not undo a previous compromise.

    No. Detection and triage are read-only. The optional defense-rules module writes only to a user-selected directory under your home or /tmp, never to system paths. The skill does not update PHP or remove malware.

    No for the normal flow. Full triage may see more logs with elevated permissions, but the skill never elevates privileges itself. If needed, it gives the command to the user to run as administrator.

    Any agent that can run shell commands, read files, and ask for confirmation. Examples: Claude Code, Codex CLI, Gemini CLI, opencode, Cursor, and VS Code Copilot when terminal access is enabled. GUI-only agents without shell access cannot run it.

    The SKILL.md workflow. On first run it clones the repository at the pinned v1.0.1 tag and downloads the Assetnote scanner and the Sairahcaz triage script from their original sources with pinned commit or hash. No MCP server or bundled scanner is included, and third-party code is not redistributed.

    No telemetry is sent. Setup downloads tools from GitHub, gist.githubusercontent.com, and PyPI. The scanner sends requests only to the target you authorize. Triage does not send network data. Reports stay on your machine.

    Reports are stored in ~/.local/state/php2shell-scans/ as JSON, Markdown, and HTML. They may contain IPs, internal paths, user names, and log fragments, so keep them private. Reports without evidence or older than 7 days are rejected by default.

    The PHP version may be hidden or unreliable. The skill reports the edge/CDN and may return INCONCLUSIVE. Confirm on the origin server or in a local/staging environment.

    The report explains the cause: 405 means the endpoint rejects POST, 413 means the body limit is too small, 429 means rate limiting, 403/406 means a WAF or access rule, and 5xx points to the server or PHP-FPM. The skill does not retry aggressively.

    No. It allows exactly one target per run. Lists, ranges, and multiple URLs are rejected by design.

    Yes, with Git Bash. WSL2 is the recommended option. Without Git Bash, the shell wrappers cannot run. On Windows, triage sees less information because PHP-FPM logs are not under /var/log.

    Passive checks usually finish in seconds. A local run includes the HTTP preflight, the active scan with a default 90-second timeout, and optional triage, so it depends on the target. A dry-run is immediate and sends no traffic.

    The skill is pinned to tag v1.0.1 for reproducibility. Updates ship as a new skill version with a new pinned tag. It does not auto-update the scanner or the scripts.

    No. It covers these specific CVEs only. A clean result does not mean the application is secure.

    It enforces safety boundaries, pins tool versions, allows one target per run, checks integrity, and turns scanner output into a consistent remediation report. Manual prompting usually skips those checks.

    New in Security