Back to skill

Security audit

Novelcraft

Security checks for vulnerabilities and agentic risk

Overview

NovelCraft appears to be a real autonomous book-writing skill, but it needs review because some templates use a fixed home directory, a fixed private-network image server, and broad subagent-control instructions.

Review or fix the templates before installing, especially every /home/felix path and the hardcoded image endpoint. Use step-by-step mode first, keep images disabled unless you trust the configured provider, run it in an isolated workspace, and avoid sensitive manuscript content until the file paths and network destination are explicitly under your control.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (2)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
templates/module-images-template.md:8
Finding

Hardcoded Cross-User Filesystem Paths in Operational Subagent Templates

Content
View full analysis

Vulnerability Details

File Location: templates/module-chapters-template.md:8-34; templates/module-concept-template.md:8-20; templates/module-images-template.md:8-26; templates/module-publication-template.md:8-33; templates/module-review-template.md:8-81; templates/module-revision-template.md:8-34; templates/module-writer-extras-template.md:8-17
Vulnerability Type: Hardcoded cross-user filesystem access
Risk Level: Medium

Vulnerable Code

Representative code from templates/module-images-template.md:8-26:

markdown
Read these files:
1. `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/project-manifest.md`
2. `/home/felix/.openclaw/workspace/novelcraft/config/module-images.md`
3. `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/00-concept/image_requests.md`

**EXTRACT IMAGE DIMENSIONS from module-images.md:**
Look for resolution configuration in the "Auflösungen pro Bildtyp" section.
If not found, use these defaults:
- cover: 768×1024
- portraits: 512×768
- world_visuals: 1024×512
- chapter_visuals: 512×512

**STEP 2: Generate Images**
Provider: MFLUX-WebUI at http://192.168.2.150:7861
Settings: low_ram=true

**STEP 3: Output Files**
Save to: `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/images/`

Additional affected path instructions include:

markdown
Save to: `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/01-drafts/chapter_{XX}_draft.md`
Copy approved chapter to: `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/02-chapters/chapter_{XX}.md`
Output: `/home/felix/.openclaw/workspace/novelcraft/Books/projects/{PROJECT}/03-final/`

Technical Analysis

Operational templates consistently use the absolute home directory /home/felix instead of resolving the workspace belonging to the invoking user. This conflicts with the localized workspace described e ...[truncated 2290 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace every /home/felix/.openclaw/... path with a workspace root supplied by the trusted OpenClaw runtime.
  2. Do not derive the home directory from user-authored project content or environment values unless the runtime validates them.
  3. Canonicalize the workspace and destination paths before every read, write, copy, or publication operation.
  4. Verify that each canonicalized path remains beneath the expected NovelCraft workspace root.
  5. Restrict project IDs to a conservative format such as ^[A-Za-z0-9][A-Za-z0-9_-]{0,63}$; reject absolute paths, separators, . components, and .. components.
  6. Reject or safely handle symlinks that resolve outside the project root.
  7. Run subagents with per-user identities and least-privilege filesystem permissions.
  8. Add automated tests that execute templates under a non-felix account and confirm that no operation touches another home directory.
  9. Update all seven affected templates together so that modules cannot resolve the same project to inconsistent locations.

T09 · Insecure Skill Coding Practices

Warning
Location
templates/module-images-template.md:18
Finding

Hardcoded Plaintext Private-Network Image Provider Conflicts with User Configuration

Content
View full analysis

Vulnerability Details

File Location: templates/module-images-template.md:18-23; related default in references/config-setup.md:102-110
Vulnerability Type: Unverified plaintext private-network endpoint and configuration bypass
Risk Level: Medium

Vulnerable Code

From templates/module-images-template.md:18-23:

markdown
- world_visuals: 1024×512
- chapter_visuals: 512×512

**STEP 2: Generate Images**
Provider: MFLUX-WebUI at http://192.168.2.150:7861
Settings: low_ram=true

Related setup example from references/config-setup.md:102-110:

markdown
## Module 2: Images (`module-images.md`)

**Create:** `~/.openclaw/workspace/novelcraft/config/module-images.md`

```markdown
# Module: Images

## Provider
- type: mcp_server
- server: mflux-webui
- endpoint: http://192.168.2.150:7861

Technical Analysis

The image template initially says to load module-images.md, but the generation step then directly specifies http://192.168.2.150:7861. This creates ambiguity over whether the user-selected provider is authoritative and may cause an agent to contact the fixed private-network service even when a different provider was configured.

Image generation can transmit prompts derived from the book's concept, characters, settings, and chapter scenes. The fixed destination is not authenticated or verified by the template, and the use of plaintext HTTP provides neither transport confidentiality nor endpoint integrity.

Because the address is fixed, this is not a general arbitrary-URL SSRF primitive in the audited text. The relevant risk is unauthorized interaction with that specific private-network address and disclosure to a service occupying or impersonating it.

Attack Path

  1. A user enables image generation and expects the provider selected in module-images.md to be used.
  2. The image subagent reads image_requests.md, which contains descriptions derived from the bo ...[truncated 1078 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hardcoded 192.168.2.150:7861 destination from the operational template.
  2. Treat the provider and endpoint in validated user configuration as the only authoritative destination.
  3. Keep image generation disabled by default, as already documented.
  4. Before the first network request, display the provider, hostname, port, transport, and categories of book data that will be transmitted.
  5. Require explicit confirmation before accessing private, loopback, link-local, or otherwise non-public network addresses.
  6. Prefer HTTPS with certificate validation for network providers. If a local provider only supports HTTP, clearly identify it as local-only and document the transport limitation.
  7. Apply destination allowlisting or explicit per-project approval rather than silently accepting arbitrary endpoints.
  8. Prevent redirects from escaping an approved destination or address class.
  9. Define authentication and response-size limits and validate returned files as expected image formats before publication.
  10. Add a test where configuration selects none, manual, or another provider and verify that no request is made to the former hardcoded address.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (60)

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The review rule 'UTF-8 Encoding ×3 (CRITICAL) — No foreign characters' imposes a language/locale restriction that is unrelated to safe UTF-8 handling and can suppress non-English or accented text. In a book-writing skill, this may cause destructive rewriting or rejection of legitimate multilingual content, creating integrity and discrimination risks for user-authored material.

Content

No source excerpt is available for this finding.

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 127)May include surrounding context.

Subagent MUST write progress header:

markdown
<!-- chapter_XX_draft.md -->
<!--
STATUS: writing
WORDS: 650

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 670)May include surrounding context.

md
| `SKILL.md` | This file — main documentation |

Self-Modification

High
Category
Rogue Agent
Confidence
95% confidence
Finding

The file explicitly instructs updating SKILL.md, which is the controlling behavior definition for the skill. Allowing a skill or its workflow docs to direct self-modification of its own instruction source creates a high-risk path for persistence, policy drift, and malicious alteration of future agent behavior.

Content

Scanner excerpt · WORKFLOW-FIXES.md (reported line 204)May include surrounding context.

md
## Implementation Tasks

### Task 1: Update SKILL.md Workflow Section
**File:** `~/.openclaw/skills/novelcraft/SKILL.md`
**Section:** "Workflow (Autonomous)" + new "Detailed Chapter Workflow"

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Destructive operations are documented without any warning about data loss or required confirmation, which is unsafe for a project-management skill handling user-created content. In this context, accidental or socially engineered invocation could irreversibly remove manuscripts or configuration state.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The contribution guidelines state 'All documentation in English', which is a natural-language locale policy constraint. Under the policy rules, forcing a specific language is a violation unless the skill offers user choice or clearly documents a justified regional/compliance need, which is not present here.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · GETTING-STARTED.md (reported line 3)May include surrounding context.

md
# NovelCraft - Getting Started Guide

**Welcome to NovelCraft!** This guide will help you create your first book - safely and step by step.

---

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · GETTING-STARTED.md (reported line 17)May include surrounding context.

which xelatex # PDF generation (optional but recommended)

If not installed:

Ubuntu/Debian: sudo apt-get install pandoc texlive-xetex

macOS: brew install pandoc --with-cask mactex

Windows: https://pandoc.org/installing.html

text

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · GETTING-STARTED.md (reported line 130)May include surrounding context.

which xelatex # PDF generation (optional but recommended)

If not installed:

Ubuntu/Debian: sudo apt-get install pandoc texlive-xetex

macOS: brew install pandoc --with-cask mactex

Windows: https://pandoc.org/installing.html

text

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The legacy natural-language examples are broad enough to be triggered by ordinary conversational requests such as asking to 'show all my books' or 'create project', which can cause the skill to activate without clear user intent to invoke NovelCraft. In an autonomous content-generation skill that can create, reconfigure, and switch projects, ambiguous activation increases the risk of unintended state changes or workflow execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README advertises fully autonomous novel generation but does not clearly warn that the skill will create, revise, and publish files across workspace directories. In an agentic environment, missing write-scope disclosure can cause users to trigger broad filesystem changes without informed consent, increasing the chance of unintended data creation or overwriting.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 22)May include surrounding context.

md
- 📁 **Draft/Approved Folders** — Quality separation
- ⚙️ **Standardized Config Schema v3.2** — Clear 3-level hierarchy
- 👥 **Target Audience Profiles** — **NEW:** Auto-configure for age groups (6-8, 8-12, 12-16, 16-25, 25+, 60+)
- 🔁 **Max 3 Revisions** — Forced rewrite if needed
- 🤖 **Subagent Architecture** — Each module runs isolated

---

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The quick-start command tells the user to start the skill but does not disclose that execution will autonomously populate project folders and generate draft/final outputs. This omission is risky because a simple invocation may immediately trigger persistent writes, revisions, and publication steps without a clear pre-execution warning.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to enumerate active subagents and forcibly kill sessions by session key, which is a privileged control-plane action unrelated to core text generation. In a multi-task or shared environment, fuzzy matching on labels could terminate unrelated work, causing denial of service, loss of user data, or disruption of other agents.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 258)May include surrounding context.

md
### Module Configs (Level 2)

Create in `~/.openclaw/workspace/novelcraft/config/`:

| Config | Key Settings |
|--------|--------------|

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The natural-language rule 'English Only — All prompts must be in English' imposes a specific language requirement on users. This is a language/locale policy constraint, and the file does not present it as an optional user choice or clearly justified region-specific compliance requirement.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · WORKFLOW-FIXES.md (reported line 141)May include surrounding context.

Locking vor dem Schreiben:

bash
# Create lock file
mkdir -p 01-drafts/.locks
lock_file="01-drafts/.locks/chapter_05.lock"

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
84% confidence
Finding

The document discloses a concrete skill path under the user's home directory (~/.openclaw/skills/novelcraft/SKILL.md), which can aid skill enumeration and targeted tampering in environments where an agent can read or modify local files. In isolation this is modest, but in an autonomous agent ecosystem it lowers the effort needed to locate high-value control files.

Content

Scanner excerpt · WORKFLOW-FIXES.md (reported line 205)May include surrounding context.

md
## Implementation Tasks

### Task 1: Update SKILL.md Workflow Section
**File:** `~/.openclaw/skills/novelcraft/SKILL.md`
**Section:** "Workflow (Autonomous)" + new "Detailed Chapter Workflow"

### Task 2: Create Workflow Helper Functions

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This manifest describes the skill as a "Fully autonomous book author" that creates complete novels from an idea, but it does not specify concrete trigger phrases, invocation boundaries, or exclusion conditions. In a manifest file, such broad capability language can create ambiguous activation scope and increase the risk of unintended invocation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The phrase "No foreign characters" in the UTF-8 review criterion is a natural-language policy issue because it effectively bans non-English or non-localized text without offering user choice or justification. This can force a specific language/locale outcome and conflicts with the requirement not to impose language restrictions without opt-in.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · handbook/markdown/chapter_05_draft.md (reported line 19)May include surrounding context.

md
### 4. Sequence is Sacred

Never write two chapters simultaneously. Continuity suffers otherwise.

### 5. Visuals = Bonus

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Line L29 lists "too many foreign characters" as a cause for rejection, which implies a policy preference against certain languages or scripts. The document does not offer a user choice of language/locale or explain a justified region-specific constraint, so this reads as a natural-language policy violation.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · handbook/markdown/chapter_05_draft.md (reported line 69)May include surrounding context.

md
| Resource | Path |
|----------|------|
| Skill Docs | `~/.openclaw/skills/novelcraft/SKILL.md` |
| Config Schema | `~/.openclaw/workspace/novelcraft/config/CONFIG-SCHEMA.md` |
| Templates | `~/.openclaw/skills/novelcraft/templates/` |
| Manifest Template | `~/.openclaw/workspace/novelcraft/config/PROJECT-MANIFEST-TEMPLATE.md` |

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/config-setup.md (reported line 364)May include surrounding context.

md
| Resource | Path |
|----------|------|
| Skill Docs | `~/.openclaw/skills/novelcraft/SKILL.md` |
| Config Schema | `~/.openclaw/workspace/novelcraft/config/CONFIG-SCHEMA.md` |
| Templates | `~/.openclaw/skills/novelcraft/templates/` |
| Manifest Template | `~/.openclaw/workspace/novelcraft/config/PROJECT-MANIFEST-TEMPLATE.md` |

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The natural-language triggers in this section are broad enough to match ordinary conversation rather than explicit tool-invocation syntax. In an autonomous content-generation skill, that increases the chance of accidental activation of actions or workflow transitions from casual user phrasing.

Content

No source excerpt is available for this finding.