Back to skill

Security audit

sitemd

Security checks for vulnerabilities and agentic risk

Overview

The skill’s website-building purpose is coherent, but it asks agents to install and run an unreviewed downloaded binary and perform production-impacting site actions without enough safety boundaries.

Review this before installing. Only use it if you trust the sitemd distribution channel, can verify the binary source and updates independently, and are comfortable giving it authority to modify and deploy websites. Treat magic links and SITEMD_TOKEN as secrets, confirm destructive or production actions before running them, and avoid using it with broad account credentials unless necessary.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Warning
Location
SKILL.md:17
Finding

Unverified Remote Binary Download and Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 17, 38-39, and 65-66
Vulnerability Type: Remote payload retrieval and execution without integrity verification
Risk Level: Medium

Vulnerable Code Snippets

Lines 17:

markdown
1. **If no binary** (`sitemd/sitemd` does not exist) — run `./sitemd/install` to download it

Lines 38-39:

markdown
- `sitemd` — Compiled binary (run `./sitemd/sitemd launch`)
- `install` — Bootstrap script (run `./sitemd/install` to download binary)

Lines 65-66:

markdown
| `sitemd_update_check` | Check for updates |
| `sitemd_update_apply` | Apply updates |

Technical Analysis

The Skill instructs the Agent to invoke a bootstrap script that downloads a compiled executable and subsequently launches that executable. It also exposes an update mechanism capable of replacing or modifying the installed software.

The submitted package contains only SKILL.md; neither the referenced sitemd/install script nor the sitemd/sitemd binary is available for inspection. The instructions do not specify:

  • The remote download endpoint.
  • A pinned binary version.
  • An expected cryptographic digest.
  • A digital signature and trusted signing key.
  • The identity of the binary publisher.
  • Integrity or authenticity verification for updates.
  • An explicit approval boundary before downloaded code is executed.

Consequently, the effective executable payload is outside the reviewed Skill and may change after the Skill itself has been audited. If the installer, hosting infrastructure, DNS resolution, release account, or update channel is compromised, an attacker could substitute a malicious binary while preserving the legitimate-looking Skill instructions.

The absence of the referenced installer from the submitted artifact also means its download behavior, command construction, destination handling, and treatment of environment variables cannot be verified.

Attack Path

  1. An Agent loads the Skill an ...[truncated 1814 chars]
Remediation
View remediation

Remediation Suggestions

  1. Include the complete installer source in the reviewed Skill package so its network, filesystem, and process behavior can be audited.
  2. Download releases only from an explicit HTTPS endpoint controlled by the verified publisher.
  3. Pin the downloaded binary to a specific immutable version rather than retrieving a mutable latest release.
  4. Publish a SHA-256 or stronger digest through a separately protected channel and verify it before execution.
  5. Digitally sign release artifacts and verify signatures against a pinned, trusted public key.
  6. Abort installation or updating if any version, digest, or signature check fails.
  7. Apply the same authenticity and integrity controls to sitemd_update_apply.
  8. Display the source URL, version, expected digest, destination, and requested action before downloading, and require explicit user approval before first execution or update.
  9. Run the installer and binary with least privilege and restrict their filesystem, network, and environment-variable access where possible.
  10. Avoid exposing SITEMD_TOKEN or unrelated credentials to installation and update processes unless they are strictly necessary.
  11. Maintain an auditable release manifest and document the publisher, update endpoint, signing procedure, and rollback process.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs the agent to transmit a live authentication magic link via third-party messaging platforms without any warning that the link is effectively a bearer credential. If an agent sends the link to the wrong recipient, uses an insecure channel, or the message is intercepted, an attacker could complete login and gain access to the site account.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill recommends creating a long-lived API key and placing it in an environment variable, but omits any warning about secret handling, rotation, scope limitation, or exposure through logs and process environments. This increases the chance that durable credentials are mishandled and later abused for unauthorized deploys or site changes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill exposes multiple high-impact operations such as deleting pages, deploying, activating sites, changing configuration, and applying updates without any safety guidance or confirmation requirements. In an autonomous-agent context, this can lead to accidental destructive actions, unauthorized production changes, or disruption if the agent invokes these tools based on ambiguous instructions.

Content

No source excerpt is available for this finding.