一键导入
patch-trust-score
Assign a confidence and risk rating to every AI-generated change. Returns safe, review-required, or high-risk verdict with specific signals.
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。
菜单
Assign a confidence and risk rating to every AI-generated change. Returns safe, review-required, or high-risk verdict with specific signals.
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。
基于 SOC 职业分类
Optimize token usage and context window discipline. Reduce costs and improve response quality through smart context management.
Unified context lifecycle for FlowDeck sessions — ingest, filter, prune, protect, summarize, and persist with telemetry.
Predict affected files, modules, APIs, tests, and DB paths before changes. Returns an impact map for human review.
Map architecture, conventions, and file structure into `.codebase/`. Use when onboarding or before deep feature work.
Plan differently when the agent has low certainty — ask for clarification or narrow scope instead of pretending full understanding.
Protect critical context from pruning during compaction. Preserve active plans, safety files, pending operations, and user intent anchors.
| name | patch-trust-score |
| description | Assign a confidence and risk rating to every AI-generated change. Returns safe, review-required, or high-risk verdict with specific signals. |
| origin | FlowDeck |
Every AI-generated patch gets a trust score before it is applied. The score gates how the change is handled: auto-apply, flag for review, or block and escalate.
| Score | Verdict | Action |
|---|---|---|
| 80–100 | safe | Auto-apply in guarded mode |
| 40–79 | review-required | Inline warning + human ACK |
| 0–39 | high-risk | Block auto-apply, route to reviewer |
| Signal | Deduction |
|---|---|
| File has prior failure history | −20 |
| Edit contains auth/crypto/payment keywords | −8 per keyword (max −30) |
| File in arch-constrained zone | −20 |
| No test coverage for file | −10 |
.codebase/FAILURES.json for prior failures on this file
b. Scan the patch content for high-risk keywords
c. Check .codebase/CONSTRAINTS.md for boundary violationsThe patch-trust hook runs automatically on every write and edit tool call. The score is logged to stdout and appended to .codebase/DECISIONS.jsonl.
For manual scoring, use the patch-trust-score skill with a file path and change description.