菜单
Define and generate the manual-first evidence pack for high-risk work. Covers 6 high-risk class definitions and the 5-artifact schema required before finalizing, pushing, or handing off.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/withkynam/vibecode-pro-max-kit --skill vc-risk-evidence-pack命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | vc-risk-evidence-pack |
| description | Define and generate the manual-first evidence pack for high-risk work. Covers 6 high-risk class definitions and the 5-artifact schema required before finalizing, pushing, or handing off. |
| argument-hint | [risk class and work description] |
| trigger_keywords | risk evidence, high-risk pack, evidence pack, risk gate, adversarial validation |
| layer | contract |
| metadata | {"author":"vibecode-pro-max-kit","version":"1.0.0"} |
Output style: Follow
process/development-protocols/communication-standards.md— answer-first, plain language, no unexplained jargon, TL;DR on long responses.
Generate and validate the manual-first evidence pack required before finalizing, pushing, or handing off high-risk implementation work.
From process/development-protocols/orchestration.md ("High-Risk Execution Handoff") and process/development-protocols/implementation-standards.md ("Risky Work Evidence Contract") — both sources agree:
verifyInstanceOwnership, BYOK secret fetch paths, MITM proxy key injection, Bright Data credential handling, or any logic that controls what a caller is allowed to access.Per task-folder artefact colocation, all artifacts go inside the selected plan's task folder (e.g. process/features/{feature}/active/{slug}_{date}/harness/ or process/general-plans/active/{slug}_{date}/harness/), so the whole pack moves with the plan as a unit. Legacy path reports/harness/ is deprecated for new writes; never write the pack to a sibling reports/ dir or any ad-hoc location. The validator script lives at .claude/skills/vc-risk-evidence-pack/scripts/validate-risk-artifacts.mjs.
risk-gate.jsonRecords the risk class, work description, and approver identity before work begins or is finalized.
{
"riskClass": "<one of the 6 classes above>",
"workDescription": "<short description of the change>",
"approver": "<person or agent that reviewed the risk classification>",
"mustStopBeforeFinalize": true
}
context-snippets.jsonRelevant code snippets with exact file and line citations for every surface the change touches in the high-risk class.
{
"snippets": [
{
"file": "packages/api/src/router/billing.ts",
"lines": "120-145",
"description": "verifyInstanceOwnership call before secret read",
"content": "<excerpt>"
}
]
}
verification.jsonDocuments every verification step taken and its result. Steps must cover both the happy path and at least one failure or boundary case.
{
"steps": [
{
"step": "<what was verified>",
"command": "<command run, if applicable>",
"result": "PASS | FAIL | SKIP",
"notes": "<observations>"
}
]
}
review-decision.jsonThe explicit reviewer decision record. Must contain APPROVE or REJECT with a written rationale — no implicit approvals.
{
"reviewer": "<name or agent>",
"decision": "APPROVE | REJECT",
"rationale": "<written reason>",
"timestamp": "<ISO 8601 date>"
}
adversarial-validation.jsonRequired when the path is high-risk or attack-sensitive (e.g. auth bypass, privilege escalation, secret exfiltration). Documents adversarial scenarios considered and whether each was ruled out.
{
"scenarios": [
{
"scenario": "<attack or misuse description>",
"ruled_out": true,
"rationale": "<why this path is not exploitable>"
}
]
}
If risk is high, do not treat the work as ready to finalize until the evidence pack exists and the reviewer decision is recorded.
If the evidence pack is missing, say so explicitly — do not proceed silently, do not imply the work is proven, and do not report DONE.
Verbatim from implementation-standards.md:
Auto-stop rule:
- if risk is
high, do not treat the work as ready to finalize until the evidence pack exists and the reviewer decision is recorded- if the evidence pack is missing, say so explicitly instead of implying the work is proven
This contract is manual-first and opt-in by risk class. It is not a default blocking hook.
Verbatim from orchestration.md:
If the risk gate says
mustStopBeforeFinalize: true, do not imply the work is fully proven until the pack exists and the reviewer decision is present. Keep this manual-first. Do not invent a blocking hook or alternate workflow owner.
Steps to confirm each artifact is complete before handoff:
risk-gate.json populated with correct risk class, work description, approver, and mustStopBeforeFinalize flagcontext-snippets.json includes all affected file:line citations for every surface touching the high-risk classverification.json documents each test step and result, covering happy path and at least one boundary/failure casereview-decision.json has explicit APPROVE or REJECT with written rationale and timestamp — no implicit approvalsadversarial-validation.json present if the path is attack-sensitive (auth bypass, privilege escalation, secret exfiltration, trust-boundary violation)Verbatim from process/development-protocols/implementation-standards.md ("Risky Work Evidence Contract"):
For high-risk work, use a manual-first evidence pack before calling the change ready for finalize, push, or human handoff.
High-risk classes include:
- auth or identity flows
- billing, payments, or credit accounting
- schema/data migrations or destructive writes
- public API or external contract changes
- deploy/runtime/container/proxy/gateway behavior
- permission, secret, or trust-boundary logic
Preferred artifact set inside the selected plan's task folder (
{slug}_{date}/harness/):
risk-gate.jsoncontext-snippets.jsonverification.jsonreview-decision.jsonadversarial-validation.jsonfor high-risk or adversarial pathsAuto-stop rule:
- if risk is
high, do not treat the work as ready to finalize until the evidence pack exists and the reviewer decision is recorded- if the evidence pack is missing, say so explicitly instead of implying the work is proven
This contract is manual-first and opt-in by risk class. It is not a default blocking hook.
Pull latest agent harness improvements from the remote kit repository. Shows a dry-run diff summary, waits for confirmation, then applies updates.
Audit agent harness health: Claude/Codex agent parity, skill registry consistency, README.md sync, and protocol file wiring. Use when agents, skills, README.md, or development-protocol files move, split, or drift.
Use when publishing harness improvements to the remote kit repo. Diffs managed files, shows what changed, bumps version, and pushes. Counterpart to vc-update (pull).
基于 SOC 职业分类
