AGENT SKILL

malware-triage

Use for defensive offline triage of an authorized malware sample or suspicious artifact, with hashes, indicators, containment notes, and safe evidence handling.

License 许可
MIT
最近更新
2026-10-31
Stars
★ 1
Forks
0

01 前言 Front Matter

name
malware-triage
description
Use for defensive offline triage of an authorized malware sample or suspicious artifact, with hashes, indicators, containment notes, and safe evidence handling.

02 资源文件 1

03 SKILL.md 正文

Baron Malware Triage

This is a narrow, optional defensive skill. It is not a malware execution framework, offensive playbook, case-management system, or replacement for Baron's security gate.

Use When

  • The owner has a local sample or suspicious artifact and asks for offline classification, indicators, metadata, or containment guidance.
  • The task needs a bounded relationship between a sample, a source repository, and a defensive remediation decision.

Hard Safety Rules

  • Confirm scope and authorization before touching the artifact.
  • Do not execute a sample, detonate it, contact its infrastructure, evade detection, establish persistence, or deliver it to a target.
  • Do not auto-install analysis tools, modify global MCP/agent configuration, or start a service.
  • Keep the sample isolated and read-only; dynamic analysis requires explicit authorization and a dedicated sandbox owned by the user.
  • Never store live credentials, secrets, payload instructions, or unbounded raw sample content in Vault memory.

Baron Contract

  1. Record SHA-256, size, file type, path scope, timestamp, and provider/version.
  2. Prefer metadata, strings, headers, signatures, and static indicators.
  3. Mark behavior as observed, inferred, or unknown; never guess intent.
  4. Redact secrets before logs, context, reports, or durable memory.
  5. Attach execution receipts to any tool-backed claim.
  6. Keep project ID and artifact identity bound through cache and retrieval.
  7. Preserve a recovery packet when analysis is blocked or interrupted.

Triage Lanes

  • Identity and hash verification.
  • File type, archive, document, script, binary, and certificate metadata.
  • Static indicators such as domains, paths, hashes, imports, and suspicious configuration values.
  • Defensive containment, rotation, patching, and monitoring recommendations.
  • Mapping an indicator to current source, dependency, test, or decision only after verifying the current repository.

Output Contract

Return:

  • SCOPE: authorized sample, excluded paths, and isolation conditions.
  • EVIDENCE: hashes, exact offsets/paths, tool receipt, and source revision.
  • FINDINGS: observed indicators, confidence, impact, and defensive action.
  • UNKNOWN: unsupported format, missing tool, or unverified behavior.
  • CONTAINMENT: safe handling and cleanup owned by the user.
  • VERIFICATION: reproducible offline checks.

Quality Gate

security-auditor independently validates security conclusions. This skill cannot approve itself, cannot invoke other subagents, and cannot turn a sample being present into proof that analysis ran. Normal coding remains usable when the optional pack or its providers are missing.

04 同仓库更多 Skills