Back to skill

Security audit

feishu-robot-interact

Security checks for vulnerabilities and agentic risk

Overview

This Feishu robot skill is purpose-aligned, but its approval controls are weak enough that unauthorized or previously rejected tasks could be treated as approved.

Install only after reviewing and fixing the approval boundary. Require verified sender identities, require FEISHU_DEVELOPER_ID for approvals, bind approvals to one pending task with expiration, fix the ask_again/status semantics, avoid truncated task keys, and protect or minimize the stored confirmation history.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/robot_interact.py:211
Finding

Developer Confirmation Responses Are Not Authenticated

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/robot_interact.py:128
Finding

Reversed Confirmation Preference Semantics Can Automatically Execute Rejected or One-Time-Approved Tasks

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/robot_interact.py:116
Finding

Truncated Task Keys Permit Authorization-Memory Collisions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • 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 (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The documented workflow would allow any message that @mentions the bot to enter task processing, despite the description implying only other robots should trigger it. In a multi-bot/group setting, this weakens the trust boundary and can let untrusted human users or arbitrary participants submit tasks, potentially causing unauthorized actions; the inconsistent confirmation-memory logic further undermines the safety control meant to gate execution.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
82% confidence
Finding

The skill describes behavior that relies on reading configuration and writing persistent confirmation memory, but it declares no explicit tool scope or permissions. In an agent environment, missing scope boundaries can lead to overbroad file/env access, making it easier for the skill to read unintended secrets or write outside the intended memory file if later implemented or interpreted permissively.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill description, prompts, confirmation replies, and operational guidance are all specified exclusively in Chinese, including required reply keywords such as "确认" and "取消". This imposes a language constraint on users and developers without documenting an opt-in choice or justified locale limitation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation explicitly recommends persistently storing raw task contents together with confirmation history in a local memory file, but provides no guidance on data minimization, retention limits, access controls, or handling of sensitive business content. In this skill's context, robot-to-robot tasks may include operational, financial, or internal data requests, so retaining plaintext task history can expose sensitive information if the host, workspace, or backups are accessed by unauthorized parties.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The module-level natural-language documentation fixes the interaction context to Chinese without indicating that users can choose another language or locale. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Tainted flow: 'path' from os.getenv (line 29, credential/environment) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/robot_interact.py (reported line 31)May include surrounding context.

python
"""保存开发者偏好记忆"""
    path = MEMORY_FILE.expanduser()
    path.parent.mkdir(parents=True, exist_ok=True)
    with open(path, "w", encoding="utf-8") as f:
        json.dump(data, f, ensure_ascii=False, indent=2)

def get_task记忆(task_type):

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The confirmation-memory logic appears inverted: the branch for '确认+不再询问' stores ask_again=True, while need_confirmation() interprets ask_again=True as requiring future confirmation. More importantly, the plain confirmation branch stores ask_again=False, which can cause future tasks of the same type to execute without re-confirmation despite the developer not opting into that behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

All headings, instructions, examples, and approval phrases are presented only in Chinese, which effectively enforces a specific language without user opt-in. The file does not indicate that the skill is region-specific or provide an alternative language option.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

This code persists task confirmation history and developer preferences to a local JSON file, which is a file-write operation affecting stored user-related data. Although the code has internal docstrings, there is no visible user-facing warning, prompt, or log message informing operators that these preferences are being saved locally.

Content

No source excerpt is available for this finding.