Evidence Graph Review
Use this skill when a reverse engineering, forensics, CTF, or authorized security case needs a
defensible handoff. It audits the existing work/<case>/ package without changing the case or
touching a target. The canonical entrypoint is the packaged Python CLI
(python "<PACK_ROOT>/scripts/run.py" case review); upstream PowerShell/Bash helpers are not invoked.
Resolve <PACK_ROOT> from the installed root SKILL.md; keep all bundled paths intact.
Scope
This skill covers:
- Scope metadata and target-activity readiness
- Evidence record structure and reproducibility fields
- References from work items and timeline entries to Evidence
- Structured Findings and Paths in report Markdown
- Optional SHA-256 verification for case-local artifacts (fail-closed path escape protection)
- A Markdown or JSON review result for a report handoff
It MUST NOT perform reconnaissance, exploitation, dynamic instrumentation, or target changes. Those actions belong to the routed analysis skill and require the case scope gate.
ACTION REQUIRED
NOW: read../ops/scope-contract.mdand../ops/evidence-finding-path.md; confirm this is a review of an existing authorized case package.NOW: confirm the case path and choose read-only review mode.NEXT: run the packaged Python reviewer:
python "<PACK_ROOT>/scripts/run.py" case review work/<case> --format markdown
ACT: resolve every error, then rerun the review before claiming a handoff is complete.
Tool dependencies
| Tool | Required | Purpose | Auto-bootstrap |
|---|---|---|---|
Python 3.10+ with the installed reverse-skill package |
Yes | Runs the read-only case review | No, use the installed package |
No network access or third-party package is required. case review is read-only with respect to
the case unless shell redirection is explicitly used to save its output.
Workflow
Phase 1: Intake
Run the review against the existing case directory:
python "<PACK_ROOT>/scripts/run.py" case review work/<case> --format markdown
Confirm that scope.md, timeline.md, workitems.md, and evidence/ are present. A non-strict
review reports scope warnings while a strict review treats warnings as handoff blockers.
建议下一步(选一个编号)
- 修复 scope.md 中的授权、范围或 network_profile 字段
- 继续检查 Evidence 记录的可复现命令和来源
- 导出当前 review 结果并附到阶段性报告
- 换 JSON 输出接入 CI 或其他审查工具
- 暂停,先确认审查范围
Phase 2: Traceability
Review the checks for:
- Evidence IDs that do not exist
- Findings without
evidence_ids - Paths without an allowed
path_typeor Evidence reference - Work items and timeline entries pointing to unknown Evidence
- Unlinked Evidence records
- Validated Findings with low confidence
An offline observation may use repro_command: n/a only when its notes field explicitly
documents the offline limitation.
Use JSON when another tool needs stable fields:
python "<PACK_ROOT>/scripts/run.py" case review work/<case> --format json
建议下一步(选一个编号)
- 补写缺失的 Evidence,并保留原始命令
- 将候选 Finding 绑定到 Evidence 后重新审查
- 为调用链或攻击链补充 P-id 和 Path 步骤
- 生成 Markdown handoff summary
- 换回 PRIMARY skill 继续分析
Phase 3: Fixity verification
When an Evidence record contains both content_hash and artifact_path, verify the case-local
artifact:
python "<PACK_ROOT>/scripts/run.py" case review work/<case> --verify-hashes --strict
The CLI accepts sha256:<64 hex characters> and checks that the artifact remains inside the case
root. A hash mismatch is a hard failure; an artifact path that escapes the case root is an error,
never a warning.
建议下一步(选一个编号)
- 修复 hash mismatch 或替换已污染的工作副本
- 为未固定的原始文件补充 SHA-256 和 artifact_path
- 继续进入报告生成阶段
- 导出 JSON 结果供 CI 保存
- 暂停并请求人工复核
Phase 4: Handoff
Use strict mode before a final report or specialist handoff:
python "<PACK_ROOT>/scripts/run.py" case review work/<case> --strict --format markdown
Save the passing output after the review finishes, ensuring the destination directory exists. Keep review output outside the case during the check so saving it does not change the records being reviewed. Review checks declared records; authorization comes from the current task.
建议下一步(选一个编号)
- 将通过的 review 结果交给
docs-generator/生成正式报告 - 回到 PRIMARY skill 补齐新的分析证据
- 归档 Markdown 和 JSON review 结果
- 暂停并请求人工复核
Language behavior contract
- Internal reasoning, tool selection, and phase control: English.
- User-visible messages, section labels, reports, and next-step menus: Chinese unless the user requests another language.
- Default bilingual labels place Chinese first and English second, separated by
/.
Bootstrap boundary
This skill has no third-party dependency. If the packaged reverse-skill CLI is unavailable, the
only allowed recovery action is the repository bootstrap path when a Python capability is
registered for the current platform. If no such capability is registered, stop and report the
missing runtime. Do not guess executable paths, download packages, or perform a manual install
from inside this skill.
Routing context
Upstream entry: any reverse, forensics, CTF, or authorized security skill that has produced a case package.
Downstream exit: docs-generator/ for a formal report, or the original PRIMARY skill when the
graph is incomplete.
Related modules: ops/evidence-finding-path.md, ops/timeline-workitem.md,
digital-forensics/ (not yet adopted), reverse-engineering/, and docs-generator/.
References
- NIST SP 800-86: Guide to Integrating Forensic Techniques into Incident Response
- SWGDE Best Practices for Computer Forensic Acquisitions
- SWGDE Best Practices for Archiving Digital and Multimedia Evidence
任务完成自检
- 我是否审查了 scope.md、timeline.md、workitems.md 和 evidence/?
- 所有 Finding 是否引用了现存 Evidence?
- 所有 Path 是否包含合法 path_type 和 Evidence 引用?
- 是否执行了 hash verification,或记录了未执行原因?
- 是否以 strict 模式重新运行并保存了 review 结果?
