Back to skill

Security audit

anchor-spec

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners report broad activation phrases and file writes during modes described as dry-run or read-only; confidence is medium, and scan coverage is incomplete.

Consider whether you need this project-anchor workflow and whether you are comfortable with the reported file writes and broad activation triggers. Scan coverage is marked partial because several referenced paths were missing, so the results do not fully establish the skill’s behavior.

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

The activation phrase "continue" is overly broad and lacks project-memory scope constraints.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

The README advertises automatic takeover on “continue” without requiring project-memory context or identifying the intended project. For a skill that maintains project files, this creates a risk of unrelated requests initiating unwanted filesystem changes, although this README alone does not implement activation.

Content

No source excerpt is available for this finding.

L066 advertises automatic activation on broad phrases “新项目” (“new project”) and “接着干” (“continue working”), without requiring an explicit anchor-spec request.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

The advertised automatic activation on “新项目” and “接着干” does not distinguish ordinary conversation from requests to manage project anchors. Because the advertised workflow maintains project files, ambiguous activation can cause changes outside the user's intended task; the documentation itself does not enforce this behavior.

Content

No source excerpt is available for this finding.

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

Broad triggers such as “continue,” “resume,” and “where were we” can activate the skill during unrelated conversations.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

Automatic model invocation is enabled, while generic phrases such as “continue” and “resume” are listed without a project-context requirement. Activation leads into instructions to create, overwrite, and delete anchor files, so a routing mistake can become an unwanted state change rather than merely an irrelevant response.

Content

No source excerpt is available for this finding.

Doctor's default dry-run performs real project archival copies.

Medium
Category
Excessive Agency
Confidence
100% confidence
Finding

The archival branch calls archive_project unconditionally, and that function performs shutil.copytree even when execute is false. Repeated supposedly dry runs create additional suffixed copies, potentially consuming substantial disk space and duplicating project data without authorization; the original project is not deleted.

Content

No source excerpt is available for this finding.

Survey mode writes and can overwrite files despite being documented as read-only.

Low
Category
Excessive Agency
Confidence
100% confidence
Finding

Survey mode creates ROOT/temp and writes a date-named report despite being described as read-only. Running it again on the same day overwrites that report, violating the no-write expectation and potentially losing existing report content.

Content

No source excerpt is available for this finding.