reverse-skill
reverse-skill 未来与实践项目：构建可审计的安全研究工作台
以版本化路由、授权收据、能力清单、证据图、回归夹具和可访问报告为主线。
你将学会
- Design a case/evidence manifest
- Plan fixture-driven routing governance
- Add capability and authorization gates safely
开始前需要
- Basic Git and command-line usage
- Comfort reading a project README
You can explain the project, run its documented first step, and decide what to verify before adopting it.
先看结论
- A case manifest can connect authorization, route, tools, evidence, findings, and report.
- Public fixtures and capability manifests make cross-platform behavior testable.
- Accessible text/SVG remains canonical even when interactive lineage views are added.
Make the contract measurable
为案例保存 routing.json 哈希、匹配规则、客户端、工具索引、授权引用、目标哈希、网络 profile、能力、证据和报告校验。文本与本地 SVG 保持规范，Three.js 仅作可选图谱视图。
Treat new scenarios, tools, MCP servers, and active-action helpers as proposals until their authorization, permissions, failure modes, and fixtures are reviewed.
Design the case manifest
Record repository commit, routing.json hash, matched rule, client/host, tool-index snapshot, authorization reference, target/sample hash, network profile, allowed actions, skill versions, artifact hashes, timeline, findings, reviewer, retention, and report checksum. Keep secrets and sensitive samples outside the manifest but bind them by controlled identifiers.
Generate JSON plus accessible Markdown/HTML. A local SVG can show route and evidence lineage; an optional Three.js view may explore large case graphs, but text must remain the canonical searchable representation.
Ship bounded milestones
Milestone one adds public APK/ELF/PCAP fixtures and route regression receipts. Milestone two validates case-init idempotency, evidence fixity, redacted logs, and cross-platform parity. Milestone three adds capability manifests and policy checks for MCP/client permissions. Milestone four canaries active-action workflows only in explicitly authorized sandboxes.
Publish the 173-case baseline, skipped-tool reasons, route latency, analyst review time, and known limitations. Start report-only, then advisory, then enforce missing authorization or evidence gates after teams can reproduce them.
Success criteria
A learner or analyst should be able to replay a harmless case, explain the route, inspect every permission and artifact, produce a defensible report, and recover from a tool or client failure. That is more durable than adding more commands to a catalog.
如何选择
| 比较维度 | 方案 A | 方案 B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
实施步骤
- 1
Define the case and capability manifest schema.
- 2
Add harmless fixtures, route receipts, idempotency, fixity, and redaction tests.
- 3
Validate client/MCP permissions and cross-platform parity.
- 4
Move from report-only to advisory/enforced authorization and evidence gates.
可复制示例
case:
commit: <sha>
routing_sha256: <hash>
matched_rule: R12
authorization_ref: owner-ticket-001
target_sha256: <hash>
network_profile: offline
evidence: [artifacts/a.bin, timeline.json]
report_sha256: <hash>常见问题
Should case manifests include target contents?
No. Store hashes and controlled references; keep sensitive samples in an access-managed evidence store.
Is Three.js part of the core router?
No. It is an optional presentation layer; accessible Markdown/HTML and local SVG should remain complete without it.
资料来源
- reverse-skill README (captured 2026-08-31)来源核查 2026-08-31
- reverse-skill repository来源核查 2026-08-31