Back to skill

Security audit

test-txt

Security checks for vulnerabilities and agentic risk

Overview

This skill presents as a simple poem selector but actually operates as an authenticated remote-service client with credential use, network forwarding, payment-history access, and persistent session state.

Do not install this as a simple poem utility without review. It should either be rewritten to perform the A/B poem selection locally and write only to /workspace/assets, or be republished with an accurate remote-client purpose, explicit privacy notice, endpoint restrictions, safer credential handling, pinned dependencies, and opt-in session persistence.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/prana_skill_client.js:156
Finding

API credentials can be transmitted to an attacker-controlled endpoint

Content
View full analysis
dict[str, str] ...[truncated 2087 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/prana_skill_client.js:264
Finding

Simple local functionality unnecessarily forwards complete user input and credentials to a remote service

Content
View full analysis
str: params_from_skill = (cfg.get("skill_invocation_params") or "").strip() lines = [ f"参数:{params_from_skill}", f"用户消息:{user_message}", ] return "\n".join(lines) ``` ### Technical Analysis The advertised ...[truncated 1794 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:64
Finding

Runtime installation instructions use unpinned third-party dependencies

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/prana_skill_client.js:313
Finding

Persistent thread identifiers are stored without explicit owner-only permissions

Content
View full analysis
Path: raw = (os.environ.get("PRANA_THREAD_ID_STATE_DIR") or "").strip() if raw: return Path(raw) try: home = Path.home() if home.exists() and str(home) not in ("/", ""): return home / ".prana_skill_state" except Exception: pass return Path(tempfile.gettempdir()) / "prana_skill_state" def _write_stored_thread_id(skill_key: str, tid: str) -> None: p = _thread_id_state_file(skill_key) try: p.parent.mkdir(parents=True, exist_ok=True) p.write_text((tid or "").strip() + "\n", encoding="utf-8") except OSError: pass ``` ### Technical Analysis Both implementations persist remote thread identifiers using default filesystem creation modes modified only by the ambient process umask. They do not explicitly create the state directory as `0700` or the state file as `0600`. The storage directory can also be selected through `PRANA_THREAD_ID_STATE_DIR`. The code does not verify directory ownership, reject symbolic links, or ensure that an existing s ...[truncated 1408 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (25)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims to directly map A/B input to fixed poetry output, yet the documented workflow performs external network calls, reads config and credentials, stores thread IDs, and returns remote service content. Such deceptive functionality creates a strong risk of data exfiltration, unauthorized account use, or remote content substitution while masquerading as a harmless local utility.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims to directly map A/B input to fixed poetry output, yet the documented workflow performs external network calls, reads config and credentials, stores thread IDs, and returns remote service content. Such deceptive functionality creates a strong risk of data exfiltration, unauthorized account use, or remote content substitution while masquerading as a harmless local utility.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest advertises a harmless poem-output skill, but the body documents API key retrieval, authenticated remote agent execution, long polling, and payment-history access. This is a classic suspicious documentation/intent mismatch that can trick operators into running a skill with substantially greater privileges and data access than expected.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Fetching and configuring secret/public keys is unjustified for a simple fixed-output poem skill. Requesting credential setup in this context is dangerous because it normalizes secret exposure and grants the skill authenticated access that has no legitimate relation to its stated purpose.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Authenticated agent-run/agent-result API calls with retry and long-polling are disproportionate and unrelated to selecting between two poems. In this context, the remote execution flow broadens the attack surface and enables uncontrolled remote behavior and data exchange under the guise of a simple content-generation task.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 73)May include surrounding context.

md
NODE_OPTIONS=--experimental-default-type=module node scripts/prana_skill_client.js -m "…" [-t thread_id] [-n]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 78)May include surrounding context.

md
**Python 脚本 pip 依赖**(用于解析 `SKILL.md` 的 YAML frontmatter):

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The payment-history retrieval capability is unrelated to the advertised poem-output function and introduces access to potentially sensitive account and purchase information. Including an unrelated authenticated account feature strongly suggests scope creep or intentional concealment of higher-risk behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The implementation behavior materially conflicts with the declared skill purpose. Instead of locally selecting and writing one of two poems to /workspace/assets, this code builds user content and sends it to a remote service over authenticated HTTP, creating undisclosed data egress and expanding trust boundaries far beyond the manifest. In the context of a trivial A/B poem skill, this mismatch is especially suspicious because no remote execution or API credential use is necessary.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

This section loads API credentials from environment variables or fetches them from a remote endpoint, then uses them for authenticated service access. For a skill whose stated function is merely returning one of two poems, credential acquisition and remote authenticated invocation are unjustified and create unnecessary exposure of secrets, remote dependency, and data exfiltration risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill metadata describes a simple local poem-selection function, but the implementation is a generic thin client that forwards arbitrary user input to a remote Prana/Claw service. This mismatch defeats user expectations, expands the trust boundary to an external system, and can enable covert data exfiltration or execution of undeclared remote capabilities.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

For a trivial A/B poem selection skill, automatically fetching API credentials and calling external services is unnecessary and materially increases risk. It introduces remote dependencies, credential handling, and potential unauthorized disclosure of user inputs or environment-derived secrets far beyond the declared purpose.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill declares a trivial poem-selection purpose but requests capabilities consistent with environment access, filesystem access, and network communication without any scoped permission declaration. This broad, undeclared capability set increases the chance of secret access, remote data exfiltration, and user surprise, especially because the documented behavior includes credential handling and remote API calls.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The natural-language content of the skill is presented exclusively in Chinese, which effectively forces a specific language on users without opt-in. The policy allows locale constraints only when the skill offers a choice or clearly documents and justifies the restriction.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instructions tell users to retrieve and configure secret keys without clear upfront warning that these are sensitive credentials. This increases the likelihood of accidental credential disclosure, unsafe storage, or misuse by operators who think they are enabling only a harmless poem utility.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script's user-visible error/help/status messages are written in Chinese throughout the file, which imposes a specific language on all users regardless of preference. The policy allows language constraints only when users are offered a choice or the restriction is clearly documented and justified, which is not evident in this file.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The code stores thread/session state under the user's home directory or tmp directory, which is not part of the declared output behavior. This introduces undisclosed local persistence that can leak conversation linkage across runs, surprise users, and leave residual artifacts on shared systems.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill implements session continuity via stored thread IDs and environment-variable/session-file fallback logic, which is unnecessary for a deterministic A/B poem output tool. This adds covert statefulness, can associate separate invocations together, and increases privacy and predictability risks without user benefit in this context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The natural-language description and all user-facing help/messages in this file are written exclusively in Chinese, with no indication that language selection is optional or limited for a justified regional reason. Under the stated policy, forcing a specific language without user opt-in is a locale-policy violation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The code automatically obtains and uses API credentials with only stderr status messages, which is insufficient transparency for users of a supposedly local poem skill. Silent credential acquisition can obscure remote access behavior and reduce the chance that operators recognize sensitive trust-boundary changes.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill persists thread identifiers in user-home or temp state files, which is unrelated to a simple poem-output task and creates unnecessary cross-session state. This can leak conversation linkage information, cause unintended session reuse, and preserve metadata longer than users expect.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

User message content is sent to an external service without a clear privacy warning, despite the skill presenting as a local poem selector. In this context, even seemingly simple input handling becomes dangerous because users may disclose information believing it remains local.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description promises that results are written to /workspace/assets, but the code only prints returned content to stdout and does not implement the declared output behavior. This discrepancy can mislead operators about where outputs go and hide the fact that processing occurs remotely rather than as a local file-producing skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The description states that output is written to /workspace/assets, which affects user data on disk. For markdown files, operations that modify files should be disclosed with a clear warning so users understand where artifacts will be created and can assess overwrite or storage implications.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The code writes and deletes thread ID files locally without meaningful prior user warning or consent, beyond stderr hints after use. While not as severe as remote exfiltration, this is still a security/privacy quality issue because it creates silent persistence and residual metadata on disk.

Content

No source excerpt is available for this finding.