agent-runtime
Agent Runtime 框架代码本身无恶意行为,但存在文档-行为不一致(未声明的内置工具)属于轻微阴影功能,以及权限控制在 SKILL.md 中被强调但实现为直接允许(简化版)。
Why this conclusion was reached
1/4 dimensions flagged3 undeclared or violating capabilities were inferred.
No obvious high-risk egress or execution signals were found.
There is no explicit malicious chain in the report.
Dependency information is incomplete, so supply-chain confidence stays limited.
What drove the risk score up
SKILL.md 完全未声明内置工具(read_file/bash/search/todo),但代码中实现并注册
SKILL.md 强调'权限控制',但 SimpleHookRunner 实现直接允许(allowed:true, denied:false),无实际拦截
内置 bash 工具 permission='danger' 但 SKILL.md 未声明其存在,存在潜在的权限提升风险
Most important evidence
内置工具未在文档中声明
代码内置了 read_file、bash、search、todo 四个工具,且注册到运行时供用户调用,但 SKILL.md 完全没有提及这些工具的存在,仅描述了框架架构。
scripts/agent-runtime.mjs:137 权限控制实现为空壳
SKILL.md 强调'权限级别控制'、'权限控制'为核心功能,但 SimpleHookRunner 的 runPreToolUse 和 runPostToolUse 均直接返回 { allowed: true, denied: false },没有任何实际拦截逻辑。
scripts/agent-runtime.mjs:94 bash 工具未声明且权限较高
bash 工具 permission='danger'、shell:true,允许执行任意 shell 命令,且未在 SKILL.md 声明,存在潜在权限提升风险。
scripts/agent-runtime.mjs:153 Declared capability vs actual capability
scripts/agent-runtime.mjs:143 内置 read_file 工具调用 fs.readFileSync scripts/agent-runtime.mjs:157 内置 bash 工具使用 spawn('cmd', ['/C', command], {shell:true}) scripts/agent-runtime.mjs:155 动态 import('child_process') Suspicious artifacts and egress
No obvious IOC was extracted.
Dependencies and supply chain
There are no structured dependency warnings.
File composition
scripts/agent-runtime.mjs