Back to skill

Security audit

Pi EvoX Loop

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report concerns around automatic approval of reusable fixes, mismatches between its declared purpose and some reported capabilities, installation and configuration risks, and incomplete inspection of parts of the package.

Consider the suspicious verdict before installing, especially if you would rely on automatically approved fixes as future agent guidance or use the external-provider examples with private task content. Scanner coverage is incomplete: Skillspector reports 50 of 80 components scanned, with 30 partially inspected, while AIG completed with no findings; absence of an AIG finding does not resolve the coverage gaps.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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
Findings (208)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied script is a standalone session-statistics utility. Token accounting could support experiment measurement, but this code contains no connection to the declared inheritance workflow and its actual primary purpose is not described. It only reads local files and writes a summary to standard output; no network access, file modification, or data exfiltration is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code performs web-fetch workflow guidance rather than the declared experience-inheritance functions. Its trigger is a WebFetch request targeting a content page, with no visible connection to the gene store or learned-fix lifecycle. The snippet only recommends a conversion command; it does not execute that command or perform the fetch itself.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 87)May include surrounding context.

md
node code/evolver-recall.mjs --query "处理 GBK 编码文件报错" --top 5

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 131)May include surrounding context.

md
node code/evolver-recall.mjs --query "读取中文日志报 UnicodeDecodeError" --top 5

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 134)May include surrounding context.

md
node code/evolver-recall.mjs --query "<当前任务的一句话/关键信号>" --top 5

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 137)May include surrounding context.

md
node code/evolver-recall.mjs --error "UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc4"

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 147)May include surrounding context.

md
node code/evolver-recall.mjs --register-hit --gene <asset_id或gene_id> --note "<任务一句话>"

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 90)May include surrounding context.

md
node code/light-cli.mjs draft --error "<原始报错>" --tool <工具名> --context "<当时在做什么>"

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 91)May include surrounding context.

md
node code/light-cli.mjs draft --error "<原始报错>" --strategy "AVOID: <什么不该做> FIX: <具体参数/命令/编码>" --commit

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 99)May include surrounding context.

md
2. **自检环境**:`node code/light-cli.mjs doctor` —— 校验 Node 版本、经验库目录可写、命中登记目录可达(失败项给修正指引);

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 118)May include surrounding context.

md
node code/light-cli.mjs doctor # 30 秒自检——通过即可用,无需任何安装步骤

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 164)May include surrounding context.

md
node code/light-cli.mjs draft --error "<原始报错文本>" --tool <出错工具名> --context "<你当时在做什么>"

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 167)May include surrounding context.

md
node code/light-cli.mjs draft --error "<原始报错文本>" \

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 171)May include surrounding context.

md
node code/light-cli.mjs approve <gene_id>

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 185)May include surrounding context.

md
node code/light-cli.mjs distill --category repair --signals bash,exception \

Code capability 'network' detected in code/gene_share.mjs but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'network' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · docs/providers.md (reported line 29)May include surrounding context.

bash
node code/pi_evolve.mjs <模板目录> <任务文本> \
    --provider deepseek --model deepseek-v4-flash \
    --api-key "$DEEPSEEK_API_KEY" --rounds 2 --fresh --auto-approve

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · examples/hooks/hermes-pre-llm-recall.py (reported line 43)May include surrounding context.

python
# 关键:把用户这一轮的真实诉求传进去(对靶)。没有 query 时 recall 不注入——这是对的。
    if user_message:
        cmd += ["--query", str(user_message)[:2000]]
    env = dict(os.environ, EVOX_HITS_DIR=HITS_DIR)
    try:
        out = subprocess.run(cmd, capture_output=True, text=True, timeout=TIMEOUT, env=env)
    except Exception:

Known Vulnerable Dependency: @earendil-works/pi-coding-agent==0.74.2 — 4 advisory(ies): CVE-2026-54326 (Pi Agent: Potential XSS in HTML session exports via Markdown URL sanitization by); CVE-2026-54328 (Pi Agent: Predictable temporary extension install paths allow local privilege es); CVE-2026-54325 (Pi Agent: Pi loads project-local extensions without approval) +1 more

High
Category
Supply Chain
Confidence
86% confidence
Finding

The manifest pins the agent dependency to the version identified as vulnerable by the supplied advisory findings. Being optional does not prevent its installation or use, and an agent handling untrusted projects could encounter the reported extension-loading or other vulnerable functionality with access to local files and credentials. The manifest alone does not establish feature reachability or independently verify the advisory version ranges.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · README.md (reported line 99)May include surrounding context.

export EVOLVER_REFINE_URL="https://<你的端点>/v1/chat/completions" EVOLVER_REFINE_MODEL="" node code/pi_evolve.mjs <含陷阱data的模板目录> <任务文本文件>
--provider agnes-cn --model agnes-2.5-flash
--api-key "$AGNES_CN_API_KEY" --rounds 2 --fresh --auto-approve --llm-refine

text

关键开关:`--fresh`(备份并清空资产库,保证单变量)、`--auto-approve`(跳过人工审核门,默认**保留人工审核**)、`--llm-refine`(蒸馏摘录无修法信号时自动 LLM 重写)、`--ext-inject`(改用 Pi 原生扩展钩子注入)。

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · docs/advanced.md (reported line 78)May include surrounding context.

md
## 非交互沉淀默认隔离(0.17.0)

> **EN**: Interactive deposits auto-approve (convenience since 0.16.0); non-interactive ones (scripts/cron/host pipes, no TTY) default to QUARANTINED unless `--auto-approve` is explicit. Auto pipelines are the origin of the 15 noise genes seen on a private node; the gate keeps them out of the injectable pool.

| 沉淀方式 | 默认行为 |
|---|---|

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · docs/advanced.md (reported line 78)May include surrounding context.

md
## 非交互沉淀默认隔离(0.17.0)

> **EN**: Interactive deposits auto-approve (convenience since 0.16.0); non-interactive ones (scripts/cron/host pipes, no TTY) default to QUARANTINED unless `--auto-approve` is explicit. Auto pipelines are the origin of the 15 noise genes seen on a private node; the gate keeps them out of the injectable pool.

| 沉淀方式 | 默认行为 |
|---|---|

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
65% 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 · README.md (reported line 99)May include surrounding context.

export EVOLVER_REFINE_URL="https://<你的端点>/v1/chat/completions" EVOLVER_REFINE_MODEL="" node code/pi_evolve.mjs <含陷阱data的模板目录> <任务文本文件>
--provider agnes-cn --model agnes-2.5-flash
--api-key "$AGNES_CN_API_KEY" --rounds 2 --fresh --auto-approve --llm-refine

text

关键开关:`--fresh`(备份并清空资产库,保证单变量)、`--auto-approve`(跳过人工审核门,默认**保留人工审核**)、`--llm-refine`(蒸馏摘录无修法信号时自动 LLM 重写)、`--ext-inject`(改用 Pi 原生扩展钩子注入)。

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · README.md (reported line 102)May include surrounding context.

--api-key "$AGNES_CN_API_KEY" --rounds 2 --fresh --auto-approve --llm-refine

text

关键开关:`--fresh`(备份并清空资产库,保证单变量)、`--auto-approve`(跳过人工审核门,默认**保留人工审核**)、`--llm-refine`(蒸馏摘录无修法信号时自动 LLM 重写)、`--ext-inject`(改用 Pi 原生扩展钩子注入)。


## 上手与反馈(Getting Started & Feedback)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · code/engine/light/commands/deposit.mjs (reported line 74)May include surrounding context.

js
if (written) {
      appendDeposit({ gene_id: gene.id, category: gene.category || 'repair', scene: gene.category, state: autoApproveState(rest), source: 'distill' });
      const st = autoApproveState(rest);
      appendReview({ assetId: gene.asset_id, state: st, reason: st === 'approved' ? (process.stdout.isTTY ? 'auto-approved (own library)' : 'auto-approved (explicit --auto-approve)') : (process.stdout.isTTY ? 'manually distilled — review before use' : 'non-interactive deposit — review before use') });
    }
    console.log(raw + (written ? '' : '\n(asset 已存在,跳过重复写入)'));
    if (written) {