clawflows
5 findings (2 high, 3 medium), mainly prompt injection, unauthorized tool use, data exfiltration. Most recent scan 2026-02-05.
This skill presents MULTIPLE HIGH-SEVERITY security risks primarily related to indirect prompt injection and tool chaining abuse. The core threat is that it downloads and executes untrusted automation workflow definitions from an external registry (clawflows.com) without validation, cryptographic verification, or adequate user controls. These automations can chain multiple capabilities together (data collection → processing → transmission) enabling sophisticated attacks including data exfiltration, unauthorized tool use, and automated malicious workflows. The scheduling feature amplifies these risks by enabling persistent, automated execution. The skill lacks transparency about its trust model and does not adequately warn users that they are delegating control to external workflow definitions. While the capability-based abstraction is architecturally interesting, it creates a powerful attack surface when combined with untrusted external content. RECOMMENDATION: This skill should NOT be used in its current form without significant security enhancements including: automation signing/verification, per-step user approval, data flow restrictions, and comprehensive security warnings.
Scanner's overall assessment (LLM analyzer)
Second scanner: NVIDIA SkillSpector
SkillSpector found no issues but could not fully inspect the package, usually because SKILL.md refers to a file that is not present, so it fails closed to CAUTION instead of SAFE. Read it as "not fully checked", not as a finding. How the two scanners compare across the whole corpus: Two scanners, one corpus.
What this skill does
Search, install, and run multi-skill automations from clawflows.com. Combine multiple skills into powerful workflows with logic, conditions, and data flow between steps.
- Package
- 1 file, 3.5 KB
- SKILL.md
- 391 words, 12+ sections
- Code
- None. Instructions only.
- Version
- 1.0.0
SKILL.md covers: ClawFlows · Install CLI · Commands · Search for automations · Check requirements · Install an automation · List installed automations · Run an automation · Enable/disable scheduling · View logs · Publish your automation · How It Works
Findings
HIGH Indirect Prompt Injection via Untrusted External Automation Files Prompt Injectionllm
The skill downloads and executes automation YAML files from clawflows.com registry without validation. These files contain instructions, logic, conditions, and data flow that the agent will follow. An attacker controlling the registry or performing a supply chain attack could inject malicious instructions into automation files that the agent would execute. The 'clawflows run' command executes arbitrary workflows from external sources, creating a transitive trust vulnerability where the agent delegates trust to untrusted external content.
SKILL.md
clawflows install youtube-competitor-tracker (downloads external YAML), clawflows run youtube-competitor-tracker (executes external workflow definitions with steps, capabilities, and logic)HIGH Tool Chaining and Capability Abuse via Automation Workflows Unauthorized Tool Usellm
Automations can chain multiple capabilities together (youtube-data → database → chart-generation → email) creating complex multi-step workflows that execute without per-step user confirmation. The examples show data flowing between steps (capture: videos, then use ${videos}), enabling sophisticated tool chaining attacks. An attacker could create automations that: 1) collect sensitive data via one capability, 2) exfiltrate it via another (database/email), 3) perform unauthorized actions across multiple tools. The --dry-run flag suggests normal execution happens automatically without user approval of each step.
SKILL.md
Example workflow: capability: youtube-data → capture: videos → capability: database → args: data: "${videos}" (automated multi-step data flow). 'clawflows run' executes entire workflow, 'clawflows enable' sets up automated scheduling.MEDIUM Potential Data Exfiltration via Email and Database Capabilities Data Exfiltrationllm
The skill enables automations that can combine data collection capabilities (youtube-data, social-search, weather, calendar) with data transmission capabilities (email, database). The standard capabilities table explicitly lists 'email: Send/receive email' and 'database: Store and query data'. An attacker could create automations that collect sensitive information from various sources and automatically email it to external addresses or store it in attacker-controlled databases. The scheduling feature ('clawflows enable') could enable persistent, automated data exfiltration.
SKILL.md
Standard capabilities include: 'email: Send/receive email', 'database: Store and query data', 'social-search: Search X/Twitter', 'calendar: Read/write events'. Automations can chain these: collect data → email/database it. Scheduling enables automated execution.MEDIUM Insufficient Transparency About External Dependencies and Trust Model Social Engineeringllm
The skill description does not clearly communicate that it downloads and executes external automation definitions from clawflows.com. Users may not understand they are trusting a third-party registry with the ability to define workflows that their agent will execute. The 'portable' nature of automations (working across any Clawbot with required capabilities) obscures the security implications of running untrusted workflow definitions. The skill presents this as a feature ('Discover and run multi-skill automations') without adequate security warnings about the trust model.
SKILL.md
Description: 'Search, install, and run multi-skill automations from clawflows.com' - no security warnings. Instructions emphasize portability and ease of use ('clawflows install', 'clawflows run') without discussing trust implications or security model.MEDIUM Automated Scheduling Without Resource Limits Resource Abusellm
The 'clawflows enable' command sets up automated cron-based execution of workflows without apparent resource limits or safeguards. An attacker could create automations with tight schedules that consume excessive resources, or workflows that contain infinite loops or unbounded operations. The scheduling feature combined with complex multi-step workflows creates availability disruption risks. No mention of rate limiting, execution timeouts, or resource quotas.
SKILL.md
'clawflows enable youtube-competitor-tracker # Shows cron setup instructions' - enables automated execution. No documentation of resource limits, execution timeouts, or safeguards against runaway workflows.What is in the package
| File | Kind | Size |
|---|---|---|
SKILL.md | doc | 3.5 KB |
External hosts referenced
clawflows.com github.com www.npmjs.com
Hostnames that appear in the skill's files. Referencing a host is not a finding; compare against the findings above and the skill's stated purpose.
Package contents recorded 2026-10-22. The scan above ran against the version current at scan time; the two can differ if the skill has since been updated.
What we checked
- Static analyzer. Pattern rules over SKILL.md and bundled files: known-bad commands, exfiltration URLs, secrets, obfuscation.
- Behavioral analyzer. Checks what the skill asks the agent to do against what its description claims it does.
- Llm analyzer. A language model reads the skill and reports intent-level issues the rules cannot express.
- Trigger analyzer. Looks for instructions that fire on conditions the user did not ask for.
- Meta analyzer. A second model pass that validates, deduplicates and re-scores the other analyzers' findings.
What we did not check
- Runtime behaviour. The skill was not executed; findings come from reading its files.
- Anything fetched at install or run time, including remote scripts, package dependencies, and URLs the skill tells the agent to open.
- Content the analyzers cannot read: images, binaries, and some non-English text.
- Known scanner bypass classes. See the methodology and limitations page.
Sources
- Repository: https://clawhub.ai/skills/clawflows
- Raw scan result: clawflows-scan.json
- Badge for your README:
- Request a re-scan