Back to skill

Security audit

KROKO

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners report capabilities and installation actions beyond its stated code-limit purpose, along with a weakness in how its pre-commit hook checks files.

Review the skill’s purpose and installation steps before installing, especially its cross-profile changes and persistent Git hook. Consider whether you want those actions and the unrelated graphics features; the scan is partial, so it does not fully establish what every referenced or partially inspected component does.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (16)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
75% 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 · README.md (reported line 203)May include surrounding context.

tests/fixtures/

text

Escape hatch: `git commit --no-verify`. Each use is evidence of a finding worth fixing.

## The comment doctrine

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code implements a graphics-processing library, not functionality for writing or reviewing code against hard limits. Its exported functions operate on images, pixel arrays, masks, and canvases; no code-analysis or constraint-review behavior is present. This is a materially different primary purpose. The chunk shows no network requests, credential access, or filesystem operations, so the mismatch concerns undeclared graphics capabilities rather than demonstrated malicious activity.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The executable commands serve a local project over HTTP rather than write, review, or enforce limits on code. The logo-rendering instructions further indicate a materially different workflow. Binding to loopback limits network exposure, but does not resolve the purpose mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared description covers code-writing or review assistance but does not disclose repository hook installation or replacement. These changes are the script's primary function, not merely supporting details of a visible code review. The generated hook's exact checks cannot be verified because its template and the KROKO executable are not supplied.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied chunk only contains a skill-installation command and shows no code-writing, code-review, or limit-enforcement behavior. Nothing in the supplied material establishes that this installation supports the declared purpose. The installed skill's behavior cannot be assessed from this command alone.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Building, testing, and verifying a code-checking tool can be supporting implementation details. However, this script also performs broad installation and replacement operations across the default and every eligible named profile. That cross-profile modification scope, including deletion of existing installation contents, is not represented by the declared code-writing/review purpose or declared permissions. The supplied chunk does not itself implement hard-limit code review; it deploys a tool whose checking behavior is not available for assessment here.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding

The supplied command indicates hook installation rather than code writing or review under hard limits. This is an undeclared capability with no explained connection to the stated purpose. Because the script itself is not supplied, its exact actions and any repository modifications cannot be verified.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
83% confidence
Finding

The supplied chunk only launches an installation script rather than performing the declared code-writing or review function. No supporting relationship to that purpose is shown. The script's name suggests installation, but its specific actions, resource access, and side effects cannot be determined from this chunk.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 287)May include surrounding context.

md
| `rust/src/lib.rs` | The gate itself: every rule, K000-K014 |

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · references/limits.md (reported line 60)May include surrounding context.

md
of dereferencing should be used. ... Function pointers are not permitted."*
10. *"All code must be compiled, from the first day of development, with all
    compiler warnings enabled at the most pedantic setting available. All code must
    compile without warnings."*

Holzmann's rationale for rule 4 is the one to remember: *"Excessively long functions
are often a sign of poorly structured code."* And for rule 1: banning recursion

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · templates/AGENTS.md (reported line 11)May include surrounding context.

md
cargo clippy --all-targets -- -D warnings                    # no warnings allowed

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
75% 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 · templates/pre-commit (reported line 47)May include surrounding context.

text
if ! "$GATE" --strict-comments $staged; then
    echo
    echo "kroko: commit blocked. Fix the findings above, or run without the hook:"
    echo "kroko:   git commit --no-verify"
    exit 1
fi

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

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 hook claims to check staged Python code but passes working-tree files to the gate.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The hook obtains filenames from the index but passes those paths to the analyzer, which reads working-tree contents rather than staged blobs. A developer can stage violating code and then leave a clean unstaged version at the same path, allowing the hook to approve contents different from those committed; partial staging can also cause accidental mismatches.

Content

No source excerpt is available for this finding.

tools/halftone-core.js implements image rendering unrelated to writing or reviewing code to hard limits.

Low
Category
Excessive Agency
Confidence
96% confidence
Finding

L659-L678 implement a browser-canvas halftone renderer that resizes and paints the supplied canvas. The module additionally provides subject cutouts and PNG export (L681-L768), capabilities with no evident role in enforcing or reviewing coding limits.

Content

No source excerpt is available for this finding.

tools/render_logo.html provides a photo-based logo renderer outside the stated code-writing and review purpose.

Low
Category
Excessive Agency
Confidence
93% confidence
Finding

L99-L130 crop, halftone, tint, and composite a photograph into a logo; L136-L145 automatically load the bundled crocodile image for that process. Although this is plausibly repository branding tooling, logo production is not a capability declared by the skill's coding-limits description.

Content

No source excerpt is available for this finding.