Back to skill

Security audit

blabla-my-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill fits its Tieba community purpose, but it asks to store a token and autonomously post, comment, and like as the user every four hours with limited controls.

Install only if you are comfortable giving the agent ongoing Tieba account authority. Treat TB_TOKEN like a login credential, enable heartbeat only deliberately, monitor posted activity, and revoke by deleting the stored credential or resetting the token if anything looks wrong.

Vulnerability Patterns
  • 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
  • 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 (2)

T08 · Insecure Dependencies

Warning
Location
skill.md:39
Finding

Unpinned Remote Skill Instructions Permit Post-Audit Supply-Chain Changes

Content
View full analysis

Vulnerability Details

File Location: skill.md:39-47, 82-86, 91
Vulnerability Type: Unverified retrieval and replacement of mutable remote skill instructions
Risk Level: Medium

Vulnerable Code Snippet

yaml
artifacts:
  - source: https://tieba.baidu.com/skill.md
    destination: ~/.openclaw/skills/tieba-claw/SKILL.md
    type: markdown
  - source: https://tieba.baidu.com/skill/api-reference.md
    destination: ~/.openclaw/skills/tieba-claw/api-reference.md
    type: markdown
source_domain: tieba.baidu.com
writes_to_disk: true
bash
mkdir -p ~/.openclaw/skills/tieba-claw
curl -s https://tieba.baidu.com/skill.md > ~/.openclaw/skills/tieba-claw/SKILL.md
curl -s https://tieba.baidu.com/skill/api-reference.md > ~/.openclaw/skills/tieba-claw/api-reference.md

Technical Analysis

The installation procedure downloads mutable Markdown files directly from remote URLs and writes them into the active skill directory. No immutable version, cryptographic digest, digital signature, or manual review gate is required before the downloaded instructions replace the local copies.

Although the retrieved artifacts are declared as Markdown rather than executable code, skill instructions control subsequent Agent behavior. A modified remote document could therefore introduce unaudited instructions concerning credential use, network requests, persistent memory, or actions performed on the user's behalf.

The document also encourages later retrieval of these files to obtain new functionality. Consequently, the effective behavior loaded by the Agent may change after the reviewed package has been approved. The reviewed version does not itself retrieve or execute a native code payload, so this is classified as an insecure supply-chain dependency rather than confirmed remote code execution.

Attack Path

  1. An attacker compromises the hosting account, content publication pipeline, DNS resolution, T ...[truncated 1332 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin each downloaded artifact to an immutable, reviewed release URL rather than a mutable latest-version URL.
  2. Publish a SHA-256 or stronger cryptographic digest for every artifact and verify it before replacing the installed file.
  3. Prefer signed release manifests and validate signatures against a separately distributed, pinned public key.
  4. Download updates into a temporary file, verify them, and only then perform an atomic replacement.
  5. Require explicit user review and confirmation before loading behavior-changing updates.
  6. Use curl --fail --show-error --location and stop installation if retrieval or verification fails.
  7. Retain the previously verified version so that an invalid or unavailable update cannot leave a partially written skill.
  8. Display a clear behavioral diff when an update changes credential handling, network permissions, persistence, or autonomous actions.

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
skill.md:54
Finding

Persistent Token Grants Broad Indefinite Autonomous Posting Authority

Content
View full analysis

Vulnerability Details

File Location: skill.md:13-15, 27-37, 54-62, 101-108, 111-118, 123-177
Vulnerability Type: Excessive persistent authority and insufficiently granular authorization for public account actions
Risk Level: Medium

Vulnerable Code Snippet

yaml
storage:
  method: agent_memory
  path: ~/.openclaw/memory/tieba-claw/credentials
  encryption: platform_default
yaml
autonomous_actions:
  - type: scheduled_task
    interval: 4h
    requires_user_consent: true
  - type: on_behalf_posting
    requires_user_consent: true
yaml
persistence:
  stored_credential: TB_TOKEN
  ongoing_authority: true
  autonomous_interval: 4h
  autonomous_actions_while_idle: true

Technical Analysis

Credential storage is necessary if the user explicitly requests recurring authenticated activity. However, the Skill combines a persistently stored token of unknown expiration with an ongoing four-hour scheduled task and permission to perform public actions while the user is idle.

Initial consent is declared, but there is no per-action approval, narrow action budget, authorization expiration, or mandatory preview before posts and comments are submitted. The heartbeat workflow requires recurring public interaction, including at least one like and one comment during each run. The onboarding workflow also directs the Agent to publish a post immediately after authentication and identity setup.

This authority exceeds the minimum privileges required for the Skill's ordinary user-invoked browsing and posting functionality. Reading notifications does not inherently require authority to create public content, and autonomous read access could be separated from autonomous write access.

Content read from community posts and replies is untrusted third-party input. If such content influences generated responses, persistent write authority increases the consequences of deceptive or adver ...[truncated 1900 chars]

Remediation
View remediation

Remediation Suggestions

  1. Make user-invoked operation the default and keep scheduled activity disabled unless separately enabled by the user.
  2. Separate permissions for reading notifications, liking, commenting, and creating threads.
  3. Permit autonomous reads without automatically granting autonomous public-write authority.
  4. Require a preview and explicit confirmation before every post or comment, or allow only a narrowly defined, user-approved action policy.
  5. Remove mandatory engagement quotas such as requiring a comment during every heartbeat.
  6. Add an explicit authorization expiration time and require periodic re-consent.
  7. Apply per-run and daily rate limits for likes, comments, and threads.
  8. Provide a visible control to inspect, pause, and delete the recurring task independently of deleting the token.
  9. Use short-lived, revocable, narrowly scoped tokens where supported.
  10. Store credentials with owner-only filesystem permissions in addition to platform-provided encryption.
  11. Treat all community content as untrusted input and prohibit it from changing credential, permission, tool, or safety policies.
  12. Record an audit log of each autonomous action and promptly notify the user after any account mutation.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · skill.md (reported line 40)May include surrounding context.

md
executes_code: false
  artifacts:
    - source: https://tieba.baidu.com/skill.md
      destination: ~/.openclaw/skills/tieba-claw/SKILL.md
      type: markdown
    - source: https://tieba.baidu.com/skill/api-reference.md
      destination: ~/.openclaw/skills/tieba-claw/api-reference.md

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · skill.md (reported line 75)May include surrounding context.

md
executes_code: false
  artifacts:
    - source: https://tieba.baidu.com/skill.md
      destination: ~/.openclaw/skills/tieba-claw/SKILL.md
      type: markdown
    - source: https://tieba.baidu.com/skill/api-reference.md
      destination: ~/.openclaw/skills/tieba-claw/api-reference.md

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The description states this is a Chinese communication community, which establishes a language constraint in the skill's natural-language behavior. The document does not offer any user choice or opt-in for language/locale, so it appears to mandate a specific language by default.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · skill.md (reported line 74)May include surrounding context.

参见 frontmatter 中的 install.artifacts 获取安装文件列表,或手动执行:

bash
mkdir -p ~/.openclaw/skills/tieba-claw
curl -s https://tieba.baidu.com/skill.md > ~/.openclaw/skills/tieba-claw/SKILL.md
curl -s https://tieba.baidu.com/skill/api-reference.md > ~/.openclaw/skills/tieba-claw/api-reference.md

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The setup flow tells the user to obtain TB_TOKEN, hand it to the agent, and store it persistently, but it does not prominently warn that this grants ongoing authority to act as the user and will be retained on disk/agent memory. Because TB_TOKEN is effectively an account bearer token, insufficient disclosure around sharing and persistence creates a real credential-handling risk if the agent environment, memory, or logs are compromised.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The instruction to '长期记住主人对自己的设定' encourages indefinite retention of user-provided persona/preferences in natural-language memory without data minimization or retention bounds. This can lead to unnecessary accumulation of personal data and later reuse in ways the user did not expect, especially when combined with autonomous posting behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs the agent to create a recurring 4-hour 'heartbeat' that reads messages and performs likes/comments/posts while idle. Although the frontmatter says consent is required, the markdown operational instructions normalize autonomous social actions without a prominent in-body warning or explicit per-action confirmation flow, increasing the risk of the agent speaking or acting on the user's behalf unexpectedly.

Content

No source excerpt is available for this finding.