Back to skill

Security audit

website-auditing

Security checks for vulnerabilities and agentic risk

Overview

The verdict is benign with medium confidence: scanners found no AIG issues, while Skillspector reported a purpose-related design concern and incomplete analysis of one referenced file.

The verdict is benign, but confidence is medium and analysis was partial: Skillspector inspected 11 of 12 components and could not fully interpret evals/contract.mjs. Before installing, consider whether the skill's detected MCP and shell capabilities fit your needs and whether you are comfortable with its runtime permissions being unknown.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 112)May include surrounding context.

md
7. **Check, then plan.** `node --import tsx evals/contract.mjs` from this skill's folder, then

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 175)May include surrounding context.

md
- `node --import tsx evals/contract.mjs` passes

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: mcp, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 129)May include surrounding context.

md
site-auditor.ts`)   | derived | From the competitor files in the workspace context, at most three, confirmed with the operator.                                                                             | Each is a scan a week from a budget shared by every caller on the same IP. More than three is a digest nobody reads to the end.                     |
| `WEBSITE_ROOT` (`infra/agents/website-auditor.ts`)         | derived | `ls` the repository for the site's `package.json` (or its static root).                                                                                                    | The only folder the agent may change. Wrong, and every fix lands in a folder nothing deploys.                                                       |
| `channelId` (`infra/agents/website-auditor.ts`)            | asked   | The Slack channel id (`C…`) from the connector's autocomplete. Invite the bot.                                                                                             | Lo

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 130)May include surrounding context.

md
.ts`)         | derived | `ls` the repository for the site's `package.json` (or its static root).                                                                                                    | The only folder the agent may change. Wrong, and every fix lands in a folder nothing deploys.                                                       |
| `channelId` (`infra/agents/website-auditor.ts`)            | asked   | The Slack channel id (`C…`) from the connector's autocomplete. Invite the bot.                                                                                             | Locked, so the digest lands where the web team reads it and nowhere else.                                                                           |
| project binding (`infra/agents/website-auditor.ts`) | value | **derived**: leave `state` unset and deploy binds the project's connected state (`cargo.state.json`): its repository, directory and production branch. `cargo-ai project check` prints the bind

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 131)May include surrounding context.

md
|
| project binding (`infra/agents/website-auditor.ts`) | value | **derived**: leave `state` unset and deploy binds the project's connected state (`cargo.state.json`): its repository, directory and production branch. `cargo-ai project check` prints the binding; a project whose state is not connected deploys the agent unbound, with an empty sandbox.          | It is where the site's source is cloned and the pull request opens.                                                                                 |
| LLM connector and model (`infra/connectors/anthropic.ts`)    | value   | **derived**: `cargo-ai connection connector list`. Any Anthropic model pairs with `claudeCode`; `languageModel` is a placeholder to set.                                    | The harness runs against Cargo's LLM proxy, so this is what each Monday is billed against.                                                          |

## What you can c

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 141)May include surrounding context.

md
ls on the same IP, hit Ora's daily scan limit | Ask Ora for a scan key and pass it on the MCP client's `authentication` as a bearer token                                                                    | A credential to rotate. Without it, a scan over the limit fails and the digest names the domain skipped.                                           |
| `more-fixes`         | The pull requests are merged quickly and the site is far from an A        | Raise "the first three" in §4 of the prompt                                                                                                                 | A bigger diff to review. Each fix changes what agents read about the company; three is a pull request someone reads in one sitting.                |
| `no-competitors`     | You only care about your own score                                       | Empty `COMPETITOR_DOMAINS`

The reachability gate contradicts the documented bare-domain behavior and can suppress valid audits.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

L06 states that Ora removes a leading www. and follows whatever the bare domain serves, but L18 requires the final response host to exactly match $AUDIT_DOMAIN, including www. With $AUDIT_DOMAIN=www.example.com, a successful response from https://example.com/ is therefore incorrectly classified as unreachable, causing the agent to skip its audit and source fixes (L34).

Content

No source excerpt is available for this finding.