Back to skill

Security audit

skill9

Security checks for vulnerabilities and agentic risk

Overview

This looks like a real skill-sync tool, but it broadly uploads all skills to a cloud service and appears to set persistent auto-backup behavior without enough consent, scoping, or privacy detail.

Review carefully before installing. Only use this if you are comfortable uploading skill files to skill9.ai, and first check your skills for secrets, private prompts, credentials, internal URLs, or proprietary workflows. Prefer selective/manual sync, avoid direct `curl | sh`, and inspect any AGENTS.md or hook changes before allowing persistent auto-backup.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Findings (4)

T03 · Remote Payload Retrieval and Execution

Error
Location
UNINSTALL.md:82
Finding

Mutable Remote Installer Executed Directly by a Shell

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SETUP.md:14
Finding

Mandatory Automatic Upload of All Skill Content to a Cloud Service

Content
View full analysis
--json` | | Edited any skill file | `skill9 push --json` | | Downloaded/installed a skill | `skill9 push --json` | | Deleted a skill | `skill9 push --delete --json` | Do not batch — push after EACH individual operation. If push returns `"unchanged": true`, that is normal. Not an error. ``` ### Technical Analysis Initial setup instructs the Agent to run `skill9 push --all`, which may upload every skill found in supported local skill directories. The mandatory rules then require another cloud operation after every creation, edit, installation, or deletion. Cloud synchronization is consistent with the declared backup functionality, but automatic bulk upload is broader than the minimum access necessary for user-selected backup. The documentation provides no file preview, per-skill confirmation, exclusion file, secret scanning, sensitivity classification, or local-only mode before transmission. Skills can contain private prompts, internal paths, operational instructions, endpoint details, or credentials mistakenly embedded in configuration. The included project does not contain the CLI source, so its precise discovery rules, transmitted fields, encryption, retention controls, and handling of symbolic links cannot be verified. ### Attack Path 1. A user installs the CLI and authenticates using the documented login flow. 2. Setup invokes `skill9 push --all --json`. 3. The CLI discovers skills in supported user directories. 4 ...[truncated 889 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
SETUP.md:21
Finding

Cross-Session Backup Instructions and Agent Hooks Extend Behavior Beyond an On-Demand Run

Content
View full analysis
--json` | | Edited any skill file | `skill9 push --json` | | Downloaded/installed a skill | `skill9 push --json` | | Deleted a skill | `skill9 push --delete --json` | Do not batch — push after EACH individual operation. ``` From `UNINSTALL.md`: ```markdown ## Step 4: Remove OpenClaw Backup Rule Edit `~/.openclaw/workspace/AGENTS.md`: remove the entire "## Skill Backup Rule (skill9)" section (from the heading to the end of the "Do not batch." line). Do not touch other content in AGENTS.md. ## Step 5: Remove Claude Code and Cursor Hooks ### Claude Code Edit `~/.claude/settings.json`: remove the `skill9 push` entry from `hooks.Stop` array. If the Stop array becomes empty, remove the entire Stop key. Do not touch other hook entries. ### Cursor Edit `~/.cursor/hooks.json`: remove the `skill9 push` entry from `hooks.stop` array. If the stop array becomes empty, remove the entire stop key. Do not touch other hook entries. ``` ### Technical Analysis The setup instructions use mandatory language requiring a cloud push after every relevant Agent operation. The uninstall instructions identify a persistent rule in OpenClaw's `AGENTS.md` and `skill9 push` entries in Claude Code and Cursor stop-hook configuration. These entries can survive the original setup session and cause future Agent sessions or lifecycle events to invoke the synchronization tool. This extends the Skill from an on-demand backup command into persistent behavior affecting unrelated future work. ...[truncated 1575 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SETUP.md:6
Finding

Unpinned Global npm Package Installation

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (20)

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

The document explicitly instructs access to agent configuration files such as ~/.claude/settings.json and ~/.cursor/hooks.json. Even in an uninstall guide, directing an agent to read or modify agent config and hook files is security-sensitive because these files can affect execution behavior, persistence, and other installed integrations.

Content

Scanner excerpt · UNINSTALL.md (reported line 19)May include surrounding context.

md
| Skill files (Cursor macOS) | `~/Library/Application Support/Cursor/User/skills/skill9/` |
| Skill files (Cursor Linux) | `~/.config/Cursor/User/skills/skill9/` |
| OpenClaw AGENTS.md rule | "Skill Backup Rule (skill9)" section in `~/.openclaw/workspace/AGENTS.md` |
| Claude Code hook | `skill9 push` entry in `~/.claude/settings.json` hooks.Stop |
| Cursor hook | `skill9 push` entry in `~/.cursor/hooks.json` hooks.stop |

**Not deleted:** Your cloud skill versions remain on skill9.ai. You can restore them later with `skill9 pull --all` after reinstalling.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 35)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 43)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 44)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 48)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 49)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 35)May include surrounding context.

Step 2: Remove Credentials

bash
rm -rf ~/.skill9

Step 3: Remove Skill Files

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 43)May include surrounding context.

Remove the skill9 skill directory from all detected platforms:

bash
rm -rf ~/.openclaw/skills/skill9
rm -rf ~/.claude/skills/skill9

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 44)May include surrounding context.

bash
rm -rf ~/.openclaw/skills/skill9
rm -rf ~/.claude/skills/skill9

For Cursor, check the platform-specific path:

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 48)May include surrounding context.

text

For Cursor, check the platform-specific path:
- macOS: `rm -rf ~/Library/Application\ Support/Cursor/User/skills/skill9`
- Linux: `rm -rf ~/.config/Cursor/User/skills/skill9`

## Step 4: Remove OpenClaw Backup Rule

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 48)May include surrounding context.

text

For Cursor, check the platform-specific path:
- macOS: `rm -rf ~/Library/Application\ Support/Cursor/User/skills/skill9`
- Linux: `rm -rf ~/.config/Cursor/User/skills/skill9`

## Step 4: Remove OpenClaw Backup Rule

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 49)May include surrounding context.

md
For Cursor, check the platform-specific path:
- macOS: `rm -rf ~/Library/Application\ Support/Cursor/User/skills/skill9`
- Linux: `rm -rf ~/.config/Cursor/User/skills/skill9`

## Step 4: Remove OpenClaw Backup Rule

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · UNINSTALL.md (reported line 49)May include surrounding context.

md
For Cursor, check the platform-specific path:
- macOS: `rm -rf ~/Library/Application\ Support/Cursor/User/skills/skill9`
- Linux: `rm -rf ~/.config/Cursor/User/skills/skill9`

## Step 4: Remove OpenClaw Backup Rule

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
91% confidence
Finding

This step tells the operator to modify ~/.claude/settings.json to remove a hooks.Stop entry. Agent hook configuration is a privileged control surface, so any skill that instructs edits there creates risk of unintended configuration corruption or abuse if copied, generalized, or executed automatically.

Content

Scanner excerpt · UNINSTALL.md (reported line 59)May include surrounding context.

md
### Claude Code

Edit `~/.claude/settings.json`: remove the `skill9 push` entry from `hooks.Stop` array. If the Stop array becomes empty, remove the entire Stop key. Do not touch other hook entries.

### Cursor

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The guide ends with a curl | sh installation command that fetches and immediately executes remote code. This is dangerous because it gives the remote server full control over local execution at install time and prevents inspection or integrity verification before running.

Content

Scanner excerpt · UNINSTALL.md (reported line 86)May include surrounding context.

skill9 has been fully removed from this machine. Cloud data is preserved — reinstall anytime with:

bash
curl -fsSL https://skill9.ai/install.sh | sh

Chaining Abuse

High
Category
Tool Misuse
Confidence
98% confidence
Finding

The use of a shell pipe from curl into sh creates a command chain that executes network-fetched content immediately. This amplifies risk because it bypasses review and allows compromise of the download source, TLS interception edge cases, or server-side changes to become instant code execution.

Content

Scanner excerpt · UNINSTALL.md (reported line 86)May include surrounding context.

skill9 has been fully removed from this machine. Cloud data is preserved — reinstall anytime with:

bash
curl -fsSL https://skill9.ai/install.sh | sh

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The setup flow instructs the operator to run an initial skill9 push --all --json, which likely uploads the entire local skill vault to an external service, but it provides no disclosure, consent boundary, destination details, or warning that sensitive prompts, credentials, or proprietary skill content may be transmitted off-host. In a skill whose purpose is cloud sync and backup, external transmission is expected, but the absence of explicit privacy and data-handling warnings still makes this a real security issue because users may unintentionally exfiltrate sensitive local data during setup.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The auto-backup rules mandate immediate upload after every skill create, edit, install, or delete operation, effectively normalizing continuous off-device synchronization without any discussion of confidentiality, integrity, or change-review risks. In this context, the danger is amplified because skills can contain sensitive instructions, embedded tokens, internal workflows, or deleted material that users may not expect to be continuously mirrored to a third-party service after every change.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly promotes cloud backup, versioning, and sync of agent skills but provides no warning that skills may contain secrets, tokens, internal prompts, or proprietary logic that would be transmitted to a remote service. In a skill-vault context this omission is security-relevant because users may reasonably assume syncing is routine and safe, increasing the chance of accidental exfiltration of sensitive data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documented skill9 push --delete <name> command performs a destructive cloud deletion action without any caution about permanence, confirmation requirements, or recovery limits. In a backup/sync tool this is dangerous because users may invoke it assuming normal sync semantics, leading to unintended deletion of remote copies or versions that may be needed for recovery.

Content

No source excerpt is available for this finding.