Back to skill

Security audit

Ziztu Micro Drama Writer

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: the checks found no strong evidence of harmful behavior, though they noted workflow and privacy questions and the review was not fully complete.

The verdict is benign, not risk-free. Consider whether you are comfortable sending scripts or other creative material to Ziztu when using that workflow. Confidence is medium, and the scan was partial: two local references could not be fully resolved and the PNG image was outside analyzer coverage.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (8)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · assets/examples/marry-me-by-friday/grid.md (reported line 31)May include surrounding context.

md
her sleeve Mia pulls the original resignation letter | evidence | Whitlock sets down a brass-cornered box: "Your grandmother left this for after you married." | Mia, Damien, Whitlock | Lobby | 3 | side-deal: revealed; forgery: partial | |
| 25 | Eleanor's Box | RELEASE | The lid lifts | A dashcam memory card and a letter in Eleanor's hand | ELEANOR (V.O.) from the letter: the card is locked, and only Vivian's driver knew the code. The driver was Chloe's father | Mia understands why Chloe could never say no to Vivian | call | Mia calls Chloe: "Your father drove her car that night." Silence. Then: "How do you know that?" | Mia, Damien, Chloe, Eleanor | Mia's Apartment | 3 | crash-cause: teased | |
| 26 | The Driver's Daughter | COMPRESS | Chloe hangs up on Mia | Vivian's tea service rattling in Chloe's hands | Vivian knows about the box: "The card is useless without the code, and the code died with your father." She moves the vote to tomorrow | Chloe pockets her father's photo from Vivi

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/ziztu.md (reported line 32)May include surrounding context.

md
|---|---|
| claude.ai and Claude Desktop | Customize, then Connectors, then **+ Add**, then **Add custom connector**. Name it Ziztu, paste `https://getziztu.com/mcp`, sign in to Ziztu and press Allow. In a chat: **+**, Connectors, switch Ziztu on. |
| Claude Code | `claude mcp add --transport http ziztu https://getziztu.com/mcp` (add `--scope user` for every project), then type `/mcp`, pick ziztu and sign in. |
| Codex (CLI, IDE extension, desktop app) | `codex mcp add ziztu --url https://getziztu.com/mcp`, then add `tool_timeout_sec = 180` under `[mcp_servers.ziztu]` in `~/.codex/config.toml` (a Director call can take longer than Codex's 60 second default), then `codex mcp login ziztu`. Check with `codex mcp list`. |
| Cursor | Add `"ziztu": { "url": "https://getziztu.com/mcp" }` inside `mcpServers` in `.cursor/mcp.json`, then sign in from Cursor's MCP settings. |
| Anything else | Step-by-step pages for each app: https://getziztu.com/connect |

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

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.

The skill defaults to a US and English-speaking target market without user opt-in.

Medium
Category
Output Handling
Confidence
85% confidence
Finding

The defaults impose a US and global English audience unless the user names another market. Preserving the user's output language does not resolve this separate locale assumption, which is neither justified by a region-specific purpose nor presented as an explicit user choice.

Content

No source excerpt is available for this finding.

The production workflow uploads scripts to Ziztu without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Step 6 instructs the agent to import the episode's Fountain script into Ziztu, transmitting user-provided or unpublished creative content to a third-party service. Although the workflow discloses paid calls and requires price confirmation, it does not warn users about the privacy implications of uploading their scripts.

Content

No source excerpt is available for this finding.

The workflow uploads revised user scripts to Ziztu without a privacy warning.

Medium
Category
Output Handling
Confidence
84% confidence
Finding

This step instructs the agent to import a revised script into a Ziztu project after a doctor pass, rather than merely listing an available tool. Although it explains version preservation, it does not warn that text supplied or developed in the conversation will be transmitted to and stored by an external service.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/ziztu.md (reported line 46)May include surrounding context.

md
**Show the price and wait for a yes.** Say what it buys in plain words ("first frames for 14 shots: 420 credits; balance 11,800"). Use `get_credits` for the balance.
3. **Spend with a cap.** Pass `max_credits` at or just above the quoted price. Above 5,000 credits the call also needs `confirm: true`, which you only send after the user has seen that price.
4. **Pilot before the batch.** Animate one shot first (`shot_ids` with one id) when the look or a character is new, and let the user see it before animating the rest.
5. **Never repeat a paid call to "retry"** without checking `list_jobs` first: the first one may have started.

Free: `create_project`, `get_project`, `list_projects`, `import_script`, `apply_plan`, `quote`, `get_credits`, `job_status`, `list_jobs`, `read_script`, `export_film`. Small metered cost: `director_chat` and `director_plan` messages. Paid with a cap: `script_to_shots`, `make_first_frames`, `animate_shots`, `make_music`, and the script editor and doctor tools.

The production workflow sends the user's brief and complete script to Ziztu without an explicit privacy disclosure.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

Steps B and C instruct the agent to transmit the story brief, character descriptions, and full Fountain script, including its title page, to a third-party service. Although the file explains Ziztu's production role and credit charges, it does not warn that unpublished creative material and any personal information in the script will leave the user's environment.

Content

No source excerpt is available for this finding.