类class io.agentscope.harness.agent.skill.curator 2.x

SkillSecurityScanner

public final class SkillSecurityScanner

源码Source io/agentscope/harness/agent/skill/curator/SkillSecurityScanner.java:46–388

技能 Security Scanner，让智能体以受控方式调用外部能力。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 scan、scanSingleFile、shouldAllow。

Static-analysis scanner for SKILL.md and skill support files. Ported from hermes-agent's tools/skills_guard.py regex library plus the trust-level × verdict install policy.

Detection categories (one or more regexes per category):

Category#EXFILTRATION — data exfiltration via curl/wget POST
Category#INJECTION — prompt-injection markers in markdown
Category#DESTRUCTIVE — rm -rf /, mkfs, dd of=/dev
Category#PERSISTENCE — crontab / systemd / shell-rc tampering
Category#NETWORK — listen sockets / reverse shells
Category#OBFUSCATION — base64 → bash, eval $(curl …), etc.

The scan returns a Verdict (SAFE / CAUTION / DANGEROUS) plus the list of findings. Callers (e.g. SkillManageTool, HarnessAgent.promoteSkill) consult #shouldAllow(TrustLevel, Verdict) to map verdict + provenance to an install decision.

This is intentionally a Java-side approximation, not a security boundary. Skills always run inside whatever sandbox the host configured (see ShellExecuteTool); the scanner is here to catch the obvious mistakes during authoring, NOT to replace the sandbox.

嵌套类摘要Nested Class Summary

类Class	类型Kind	说明Description
`Severity`	枚举enum	定义严重程度相关取值的枚举。 A enum in io.agentscope.harness.agent.skill.curator that defines enum values used when components lets agents invoke external capabilities under control.
`Category`	枚举enum	定义类别相关取值的枚举。 A enum in io.agentscope.harness.agent.skill.curator that defines enum values used when components lets agents invoke external capabilities under control.
`Verdict`	枚举enum	定义Verdict相关取值的枚举。 A enum in io.agentscope.harness.agent.skill.curator that defines enum values used when components lets agents invoke external capabilities under control.
`TrustLevel`	枚举enum	定义Trust 级别相关取值的枚举。 Where the skill came from.
`Finding`	记录record	承载 Finding数据的记录类型。 A record in io.agentscope.harness.agent.skill.curator that carries data used when components lets agents invoke external capabilities under control.
`ScanResult`	记录record	承载 Scan 结果数据的记录类型。 A record in io.agentscope.harness.agent.skill.curator that carries data used when components lets agents invoke external capabilities under control.

方法摘要Method Summary

方法Method	说明Description
`public static` `ScanResult scan(String skillName, String skillMd, Map<String, String> resources)` static	执行 `scan` 方法。 Scan a complete skill: SKILL.md + every support file.
`public static` `ScanResult scanSingleFile(String relPath, String content)` static	执行 `scanSingleFile` 方法。 Scan a single file (used after `write_file` / `patch`).
`public static boolean shouldAllow(TrustLevel trust, Verdict verdict)` static	执行 `shouldAllow` 方法。 Map (trust × verdict) to an install decision.

方法详情Method Detail

scan

public static ScanResult scan(String skillName, String skillMd, Map<String, String> resources)

static

执行 scan 方法。

Scan a complete skill: SKILL.md + every support file.

参数Parameters

名称Name	类型Type	说明Description
`skillName`	String	技能名称（类型 `String`）。The skill name.
`skillMd`	String	技能 Md（类型 `String`）。The skill md.
`resources`	Map<String, String>	resources（类型 `Map`）。The resources.

返回Returns ScanResult

返回ScanResult。Returns scan result.

源码Source io/agentscope/harness/agent/skill/curator/SkillSecurityScanner.java:262–277

scanSingleFile

public static ScanResult scanSingleFile(String relPath, String content)

static

执行 scanSingleFile 方法。

Scan a single file (used after write_file / patch).

参数Parameters

名称Name	类型Type	说明Description
`relPath`	String	rel 路径（类型 `String`）。The rel path.
`content`	String	内容（类型 `String`）。The content.

返回Returns ScanResult

返回ScanResult。Returns scan result.

源码Source io/agentscope/harness/agent/skill/curator/SkillSecurityScanner.java:280–284

shouldAllow

public static boolean shouldAllow(TrustLevel trust, Verdict verdict)

static

执行 shouldAllow 方法。

Map (trust × verdict) to an install decision.

参数Parameters

名称Name	类型Type	说明Description
`trust`	TrustLevel	trust（类型 `TrustLevel`）。The trust.
`verdict`	Verdict	verdict（类型 `Verdict`）。The verdict.

返回Returns boolean

返回布尔值，表示判断结果。`true` if writing the skill should proceed; `false` if it must be blocked / rolled back.

源码Source io/agentscope/harness/agent/skill/curator/SkillSecurityScanner.java:293–304

嵌套类详情Nested Class Detail

Severity

public enum Severity

定义严重程度相关取值的枚举。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 LOW、MEDIUM、HIGH、CRITICAL。

A enum in io.agentscope.harness.agent.skill.curator that defines enum values used when components lets agents invoke external capabilities under control. Declared in io.agentscope.harness.agent.skill.curator.

Verdict

public enum Verdict

定义Verdict相关取值的枚举。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 SAFE、CAUTION、DANGEROUS。

TrustLevel

public enum TrustLevel

定义Trust 级别相关取值的枚举。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 BUILTIN、TRUSTED、COMMUNITY、AGENT_CREATED。

Where the skill came from.

Finding

public record Finding( String patternId, Severity severity, Category category, String file, int line, String matchText, String description)

承载 Finding数据的记录类型。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 patternId、severity、category、file、line、matchText。

A record in io.agentscope.harness.agent.skill.curator that carries data used when components lets agents invoke external capabilities under control. Declared in io.agentscope.harness.agent.skill.curator.

ScanResult

public record ScanResult(Verdict verdict, List<Finding> findings, String reportText)

承载 Scan 结果数据的记录类型。位于 io.agentscope.harness.agent.skill.curator 包。核心成员包括 verdict、findings、reportText。