技能 mantis-calibrate

📦

mantis-calibrate

Name: mantis-calibrate
Author: google

内容修订 r1 高风险 ⚙️ 外部命令📁 文件系统访问

Calibrate Mantis Security Findings

Security findings need consistent priorities grounded in exploit evidence and architectural context. This skill applies a risk matrix and documented caps to existing Mantis findings.

支持: Claude Codex Code(CC)

⚠️ 38 差

安装前请停下并请求确认。

请审查安装计划，并在更改文件前获得用户明确同意。

使用我的智能体安装

将此请求复制给你的智能体，其中包含权威 Skill 页面和清单。

智能体请求

Review the Skillstore skill "mantis-calibrate" from https://skillstore.io/skills/google-mantis-calibrate.md and its manifest at https://skillstore.io/api/skills/google-mantis-calibrate/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

你的智能体仍应展示其计划，并按安全策略请求必要确认。

Agent 可读资源

当 AI Agent、爬虫或脚本需要干净上下文，而不是读取完整页面时，请使用这些链接。

Markdown 详情 GET /skills/google-mantis-calibrate.md 已签名 manifest GET /api/skills/google-mantis-calibrate/manifest 已签名 lockfile GET /api/skills/google-mantis-calibrate/lockfile

测试它

正在使用“mantis-calibrate”。 Impact 4, likelihood 3, internal exposure, viable production finding, and no additional caps.

预期结果:

Risk score: 5.6. Priority: MEDIUM. Internal exposure applies a 0.8 multiplier to the combined impact and likelihood.

正在使用“mantis-calibrate”。 A pinned finding has failed reproduction, but its discovery snapshot does not match the active snapshot.

预期结果:

STALE_EVIDENCE: reproduction evidence does not match the active snapshot.
Reproduction failure rule: UNKNOWN. Do not force LOW based on stale evidence.
Keep the conservative calculated score and record the provenance warning.

正在使用“mantis-calibrate”。 A viable finding has an unresolved physical-access prerequisite during calibration.

预期结果:

Calibration is provisional. Mark the physical-access rule UNKNOWN, retain the calculated priority, and request evidence resolving the prerequisite.

安全审计

高风险

v1 • 2026/10/4 打开版本化报告

All 203 static detections are false positives involving Markdown syntax, workflow metadata, or descriptions of vulnerability evidence. One semantic finding identifies helper reuse without integrity verification, enabling code execution if an attacker controls the workspace helper. No evidence found of deliberate prompt injection or data exfiltration in the supplied files.

已扫描文件

914

分析行数

复核项

已忽略误报

已确认安全问题 (1)

高

SKILL.md:554-560 SKILL.md:562-572

Workspace Helper Reuse Without Integrity Verification

The skill executes an existing workspace helper after checking only "# MANTIS_HELPER_VERSION = 2". An attacker controlling that helper can retain the comment and execute arbitrary code with agent permissions.

The instructions explicitly combine helper execution with a first-line version check, without requiring body verification. Exploitation requires attacker control of the workspace helper; no malicious helper is supplied.

风险因素

⚙️ 外部命令 (50)

📁 文件系统访问 (7)

SKILL.md:29 SKILL.md:71 SKILL.md:77 SKILL.md:95 SKILL.md:97 SKILL.md:117 SKILL.md:92

审计者: codex

分享与引用此报告

分享版本化评估报告、中性徽章、嵌入卡片和引用。Skillstore 只报告证据，不判断该 Skill 是否安全。

打开版本化报告

复制报告链接

https://skillstore.io/skills/google-mantis-calibrate/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/google-mantis-calibrate/security.svg)](https://skillstore.io/skills/google-mantis-calibrate?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/google-mantis-calibrate?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/google-mantis-calibrate/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/google-mantis-calibrate.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

学术引用 (APA · BibTeX · CFF)

APA 引用

google. (2026). mantis-calibrate security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/google-mantis-calibrate/audits/1

BibTeX 引用

@techreport{google-google-mantis-calibrate-2026,
  author = {google},
  title = {mantis-calibrate security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/google-mantis-calibrate/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "mantis-calibrate security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "google"
date-released: "2026-10-04"
url: "https://skillstore.io/skills/google-mantis-calibrate/audits/1"
identifiers:
  - type: other
    value: "skillstore:google-mantis-calibrate:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 评分

评分依据证据置信度: 中

架构

可维护性

内容

社区

规范符合性

你能构建什么

Prioritize Existing Vulnerabilities

Calibrate completed findings with exploit evidence, exposure, and blast radius to support remediation planning.

Review Project Threat Context

Apply project threat models and asset criticality when assessing findings across internal components and public interfaces.

Track Evidence Across Pipeline Passes

Record snapshot provenance, stale evidence, and unresolved calibration rules in finding histories for later review.

试试这些提示

Review Calibration Inputs

Review my existing Mantis findings without modifying files. Identify missing reproduction, viability, and threat-model information needed for calibration.

Calibrate a Finding

Calibrate finding [finding_id] using its evidence and threat model. Explain the score, priority, and applicable caps before updating the file.

Audit Unresolved Rules

Review calibrated findings for UNKNOWN rules and stale evidence. Explain which priorities remain provisional and what evidence would resolve each warning.

Calibrate a Pinned Batch

Calibrate findings using [snapshot_root] and [state_root]. Verify pass alignment, preserve conservative scores for stale evidence, and review helper integrity before execution.

最佳实践

Provide reproduction results, production viability, attacker prerequisites, and a current threat model before calibration.
Use matching pinned snapshots and review every stale-evidence or UNKNOWN warning before accepting priorities.
Back up finding files and verify complete helper contents before executing generated or reused scripts.

避免

Use calibration to discover new vulnerabilities or implement patches.
Treat missing reproduction or threat-model information as proof that a finding is harmless.
Execute existing workspace helpers solely because their version comment matches.

常见问题

What does this skill calculate?

It calculates impact, likelihood, context-adjusted risk scores, and priorities for eligible existing Mantis findings.

Does it find vulnerabilities or fix code?

No. It calibrates findings already produced by earlier pipeline stages and updates their reporting fields.

What happens when evidence fields are missing?

Missing viability defaults to CONDITIONAL_VIABLE, and missing reproduction defaults to not_attempted. Applicable rules can reduce priority, except where stale-evidence safeguards apply.

Does calibration require a pinned snapshot?

No. Without active snapshot state, it uses default mode without provenance checks. Pinned and HALT modes enable conservative stale-evidence handling.

Which files can it change?

It updates workspace finding files and creates a reusable calibration helper. Pinned target source must remain unchanged.

Is the helper version check sufficient for security?

No. A matching comment does not establish integrity. Review complete helper contents or regenerate from trusted templates before execution.