Back to skill

Security audit

Agentsmd Skill

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because scanners report automatic project-file writes and skill updates that do not match its documented purpose and approval expectations.

Consider the suspicious verdict before installing. Review whether automatic lifecycle logging and self-updates fit your needs, and whether you are comfortable with unpinned installation commands and persistent project records. Scanner coverage was marked partial because several local references could not be resolved, so the evidence is incomplete; the separate AIG scan completed without findings, but that does not establish that the skill is safe.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (23)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied command performs a skill installation update, not the declared lesson-distillation or project-documentation workflow. Self-updating the installed skill is an undeclared capability involving a different resource scope. The snippet also contains no approval gate before potentially changing files, despite the description requiring confirmation for all file changes.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Progress and handover bookkeeping is related to the declared synchronization features, so logging alone would not establish a mismatch. However, this code performs those writes automatically whenever configuration is enabled and a qualifying lifecycle event arrives; it contains no approval gate enforcing the explicit requirement that all file changes receive user confirmation first. Its primary function is lifecycle journaling and handover prompting, not collecting lessons, verifying commands, or updating AGENTS.md. The code limits itself to the project and records metadata rather than prompt or transcript contents, but those safeguards do not resolve the approval and trigger discrepancies.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The command updates the installed skill repository, a capability not described by the declared project-documentation workflow. It can contact a Git remote and modify local installation files. The supplied chunk contains no confirmation mechanism, despite the description requiring approval before all file changes.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The command performs skill maintenance, which is materially different from the declared project-documentation workflow. The supplied chunk does not collect lessons, merge project rules, organize documents, or request approval before updating the skill. This assessment is limited to the provided command.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied command installs or adds a skill; it does not collect lessons, merge project rules, verify commands, or update project records. Although it appears in a README and may be installation guidance, its executable behavior is materially different from the declared purpose. The snippet does not establish whether the installed skill implements that purpose or obtains approval before changes.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The provided command performs skill installation, not the declared AGENTS.md or documentation workflow. Adding an external skill is a distinct capability, not merely an implementation detail of editing project guidelines. No lesson collection, command verification, document merging, or user-confirmation gate appears in this chunk. This assessment is limited to the supplied command and does not infer what the installed skill subsequently does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied shell snippet performs skill-package maintenance, which is not among the declared capabilities. It does not collect lessons, update AGENTS.md, or organize project records. No user-confirmation step is shown, although the surrounding workflow cannot be determined from this snippet.

Content

No source excerpt is available for this finding.

The instructions mandate specific output languages without user opt-in or a documented justification.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Line 3 requires Simplified Chinese for collaboration documents and release notes, and English for commit subjects. These are explicit language requirements, not clarity idioms, and the file neither offers a user choice nor explains why those restrictions are necessary.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

Although this is not an MCP server declaration, the installation command invokes an unpinned npm CLI through npx, which can download and execute a changing package release with the user's privileges. A compromised release could execute arbitrary code, and the unpinned skill repository adds another mutable trust dependency.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The update command runs an unpinned npm CLI through npx, allowing a subsequently compromised release to execute with the user's privileges. Updating a persistent agent skill also introduces new instructions into future sessions, so unaudited updates can affect later agent behavior as well as local files.

Content

No source excerpt is available for this finding.

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.

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 · SKILL.md (reported line 30)May include surrounding context.

md
1. **Concise**: every rule is one line, imperative, directly executable. No narrative, no explanatory paragraphs.
2. **Compliant**: structure follows the open [agents.md](https://agents.md/) spec (see `references/agents-md-spec.md`) — pure Markdown, standard sections, written for AI coding agents.
3. **No bloat**: `AGENTS.md` is organic consolidation, not an append-only log. Hard cap **120 lines** (excluding blank lines); near the cap you must delete before adding. Content that still has long-term value but fails the root-file admission bar moves to `docs/` with a one-line reference (see Step 4). Rules that lost generality, were superseded, or duplicate others get deleted or merged.
4. **Confirm before writing**: every write — updating `AGENTS.md`, agent memory, progress/status files, moving collected files — must be shown to the user and approved first. Without approval, propose only; touch nothing.

## Workflow

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Lifecycle hooks modify project files without the user-confirmation step promised by the manifest.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

When enabled, the hook automatically creates directories, overwrites runtime state, appends events, and creates handover files without presenting the write plan and obtaining the confirmation promised by the skill metadata. The enabled flag and project-directory check limit activation and scope, but do not demonstrate informed approval for these writes; routine lifecycle events can therefore produce unauthorized persistent project changes.

Content

No source excerpt is available for this finding.

The documented template convention requires English section headings without user opt-in.

Low
Category
Output Handling
Confidence
93% confidence
Finding

The language convention explicitly states that section headings must remain in English, even though entry text follows the project's primary language. This imposes a specific output language for headings without offering a user choice or explaining why English headings are necessary.

Content

No source excerpt is available for this finding.