Back to skill

Security audit

session-retrospective

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: the completed AIG scan reported no findings, while Skillspector found mostly purpose-related signals but only partially inspected one referenced script.

The verdict is benign, but confidence is medium because one referenced script was only partially inspected. Consider that uncertainty when deciding whether to install; the reported scans do not establish that the skill is risk-free.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.en.md (reported line 30)May include surrounding context.

md
report-style retro skills: this skill is the machine-aggregable
  signal-data layer, not a report for humans. Smoke-tested on a real
  session across both the main path and the corrupted-line exception.
  Triggers on "session review", "log session feedback", "retrospective",
  "fix from the feedback log", "start fixing", "回顾一下这个会话",
  "记录会话反馈", "会话复盘", "开始修复".
---

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.en.md (reported line 68)May include surrounding context.

md
## When to Use

- Engineer says "review this session", "log session feedback", or "session review"
- End of a working session on any workflow task
- After resolving a problem that took multiple back-and-forth exchanges
- Engineer says "fix from the feedback log", "process the feedback", or "start fixing" → go straight to [Fix Mode](#fix-mode-optional-second-phase)

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.en.md (reported line 289)May include surrounding context.

md
## Language Support

This skill supports both Chinese and English interaction. Triggers include "session review", "log session feedback", "retrospective", "start fixing", "fix from the feedback log", "回顾一下这个会话", "记录会话反馈", "会话复盘", "开始修复". Write notes in the session's primary language.

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.md (reported line 14)May include surrounding context.

md
- 反馈日志
  - 开发工具
description: >
  会话结束时给工作流做一次只读复盘:只提取真实发生的系统性问题——需纠正的错误答案、跳过的闸门、错误路由、MCP 错误、skill/文档缺口、多轮纠正——并以字段枚举校验的 JSONL 条目追加到本地反馈日志:同一问题复发即规则缺口,优先级自动升级;无事记一条 success,绝不编造。说一句"开始修复"即进入修复模式:按复发与优先级排序逐项出修复预案,一项一确认后执行,每项回写闭环条目保持 append-only 可审计,且修复模式只凭日志即可跨会话运行。内置零依赖校验脚本(Node 18+):写入后与修复前机器校验日志健康度与复发分组。与复盘报告类技能互补:本技能是机器可聚合的信号数据层,不是给人读的报告。已对真实会话完成主路径与损坏行异常双场景冒烟。触发词:"回顾一下这个会话"、"记录会话反馈"、"会话复盘"、"根据反馈日志修复"、"开始修复"、"session review"、"log session feedback"、"retrospective"。
---

# 会话回顾

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.md (reported line 236)May include surrounding context.

md
## 语言支持

本技能支持中英文交互。触发词包括"回顾一下这个会话"、"记录会话反馈"、"会话复盘"、"根据反馈日志修复"、"开始修复"、"session review"、"log session feedback"、"retrospective"、"start fixing"。notes 用会话的主要语言记录。

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

md
node scripts/validate-feedback.mjs metrics/workflow-feedback.jsonl

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 162)May include surrounding context.

md
- [ ] 追加后已重读日志,确认新增行落盘且通过校验(或 `validate-feedback.mjs` 退出码为 0)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 189)May include surrounding context.

md
1. **读取日志并全量校验**:加载反馈日志,按「自动检查」逐行校验并统计复发计数;损坏行按 SR-E04 跳过并报告。可运行 `node scripts/validate-feedback.mjs <日志路径>` 一步完成(含修复排序建议)。

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.en.md (reported line 281)May include surrounding context.

md
**Q: The log keeps growing — what then?**
A: This skill only appends. When size hurts readability, the engineer decides an archiving/cleanup strategy; the recurring marks help identify which patterns have already been absorbed into rules and can be archived.

**Q: Will Fix Mode touch my files without confirmation?**
A: No. It proposes a fix per item and applies one confirmation at a time; every fix appends a closure entry pointing at the original entry's timestamp, keeping the log append-only and auditable. Fix Mode runs from the log alone, across sessions.

**Q: How does this relate to experience-promotion or self-improvement skills?**