Back to skill

Security audit

Tilda Publisher

Security checks for vulnerabilities and agentic risk

Overview

This Tilda publishing skill is purpose-aligned, but it needs review because it stores account credentials and reusable browser sessions in the workspace while making live website changes.

Review before installing. Use a dedicated low-privilege Tilda account if possible, avoid storing your main password in the workspace, protect or delete .env and tilda-session.json, ensure they cannot be committed or shared, and require explicit confirmation of the target project, page, and publish/edit action before running it on a live site.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:79
Finding

Plaintext Tilda credentials stored in the workspace

Content
View full analysis
l.startsWith(`${key}=`)); const line = `${key}=${value}`; if (idx >= 0) { lines[idx] = line; } else { lines.push(line); } } fs.writeFileSync(envPath, lines.join('\n') + '\n'); console.log('Settings saved to .env'); } saveToEnv({ TILDA_EMAIL: emailFromUser, TILDA_PASSWORD: passwordFromUser, TILDA_PROJECT_NAME: projectFromUser || '', }); ``` ### Technical Analysis The onboarding procedure writes the user's Tilda email and raw password to `.env` in the current workspace. The write operation does not set restrictive permissions, validate file ownership, encrypt the values, or ensure that the file is excluded from source control and backups. The effective permissions therefore depend on the process umask and any permissions already assigned to an existing `.env` file. Other local users, workspace tools, backup systems, indexing services, or processes running under the same account may be able to read the credential. The file could also be committed or uploaded accidentally. The README states that the password is stored locally, but local storage alone does not adequately protect a reusable account password. ### Attack Path 1. A user completes onboarding and supplies a Tilda email and password. 2. The Skill writes both values in plaintext to `/.env`. 3. An attacker obtains read access through another process, shared workspace access, ...[truncated 1005 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:477
Finding

Reusable authenticated browser session stored in plaintext

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:115
Finding

Unpinned Playwright package installed and executed at runtime

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (12)

Credential Access

High
Category
Privilege Escalation
Confidence
98% confidence
Finding

The skill explicitly asks the user for their Tilda account password and states it will be stored in the workspace .env file. Storing raw credentials in plaintext on disk materially increases the risk of credential disclosure through other tools, workspace access, backups, logs, or accidental commits.

Content

Scanner excerpt · SKILL.md (reported line 64)May include surrounding context.

md
**2. Password:**
> "What is your Tilda account password?
> ⚠️ It will be stored locally in the workspace .env file — never sent anywhere."

**3. Project name (optional):**
> "What is the name of the Tilda project where articles should be published?

Credential Access

High
Category
Privilege Escalation
Confidence
98% confidence
Finding

This section introduces the mechanism for saving credentials to .env, reinforcing a design that stores authentication material in plaintext within the project workspace. Because the skill automates login to a live CMS, compromise of that file could directly enable unauthorized site access and content manipulation.

Content

Scanner excerpt · SKILL.md (reported line 77)May include surrounding context.

  • ✅ "Connection successful. Skill is ready!"
  • ❌ "Login failed. Please double-check your email and password."

Step 3 — Save to .env

javascript
const fs = require('fs');

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

The code uses path.join(process.cwd(), '.env') to write secrets into the current workspace, which is a risky location because workspace files are often readable by other tools, included in archives, or accidentally version-controlled. In this skill's context, that would expose credentials capable of logging into and modifying a user's Tilda site.

Content

Scanner excerpt · SKILL.md (reported line 84)May include surrounding context.

md
const path = require('path');

function saveToEnv(vars) {
  const envPath = path.join(process.cwd(), '.env');
  let existing = '';

  if (fs.existsSync(envPath)) {

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

fs.writeFileSync(envPath, ...) persists the Tilda email and password in plaintext, creating a durable credential store with no encryption, access controls, or lifecycle management. Given the skill's ability to publish and edit live content, theft of this file could lead to full account compromise and unauthorized website changes.

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

md
}

  fs.writeFileSync(envPath, lines.join('\n') + '\n');
  console.log('Settings saved to .env');
}

saveToEnv({

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README describes creating pages, editing content, and publishing live URLs but does not clearly warn users that these actions affect production content and may overwrite or expose changes immediately. In a browser-automation skill with CMS credentials, the lack of an explicit destructive-action warning makes accidental live-site modifications more likely, especially when combined with broad invocation and autonomous operation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The invocation guidance is broad enough that an agent may trigger this skill whenever a user mentions publishing or editing a Tilda article, without requiring strong confirmation of scope, target site, or publication intent. Because the skill performs live browser automation against a production CMS and can publish content, overly loose triggering increases the chance of unintended or premature actions on real web properties.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

For markdown files, warnings should disclose behaviours that affect user data or system integrity. The document explains that it creates, edits, and publishes pages through browser automation, but it does not prominently warn that running the skill will change live website content and may publish drafts or overwrite existing page content.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes a browser-automation skill for creating, editing, and publishing Tilda articles. However, the onboarding flow directs the agent to invoke child_process.execSync to run package installation commands, which is a general code-execution capability not justified by the publishing task itself.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The onboarding flow executes npx playwright install chromium without pinning a version, so behavior and downloaded code can change over time and may pull an unexpected package version. In a skill that also handles credentials and browser automation, unpinned tool installation increases supply-chain risk and harms reproducibility.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The install instruction npx playwright install chromium references an unpinned package resolver path, which can fetch whatever version is current at execution time. That creates a supply-chain and integrity risk, especially for an automation skill that may run in privileged local environments with stored secrets.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Because this is a markdown skill file, vague-trigger review applies. Phrases such as "add a post to my site," "change the text on the page [title]," "publish the draft [title]," and especially "release the page that was on hold" are not tightly scoped to Tilda and could match common editing or publishing requests outside this skill's intended context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The self-healing flow captures full-page screenshots, enumerates large sets of DOM elements, and persists learned selector state to disk. On authenticated CMS pages, those artifacts can contain sensitive content, administrative metadata, URLs, or session-adjacent information far beyond what is needed to publish an article, creating unnecessary data exposure and retention risk.

Content

No source excerpt is available for this finding.