Back to skill

Security audit

Copy My Profile

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent profile-export purpose, but it broadly reads AI memory/config files and encourages persistent cross-tool imports without enough scoping or review controls.

Install only if you intentionally want to export AI-memory content. Review the generated profile carefully before sharing or importing it, remove secrets and project-specific instructions, avoid importing repository-derived rules into global memory, and prefer manual diffs or project-scoped storage for target tools.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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
Findings (1)

T02 · Agent Memory Poisoning

Warning
Location
SKILL.md:50
Finding

Untrusted Project Instructions Can Be Exported into Persistent Global Agent Memory

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (18)

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

These steps operationalize the exfiltration by directing the agent to search memory, read multiple memory files and recent logs, and then reproduce the extracted contents both in a file and inline. Reading recent logs and broad memory locations substantially increases the chance of collecting sensitive data beyond what the user expected, including instructions, personal details, project context, or embedded secrets.

Content

No source excerpt is available for this finding.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

The skill instructs access to agent configuration directories such as ~/.continue/config.json, which may contain system prompts, tokens, internal paths, or other sensitive configuration data not intended for profile export. Expanding collection into home-directory agent config increases privilege and sensitivity beyond normal workspace-scoped memory extraction.

Content

Scanner excerpt · SKILL.md (reported line 65)May include surrounding context.

md
- **Codex CLI**: `AGENTS.md`, `~/.codex/instructions.md`
   - **Cursor**: project `.cursorrules`, `.cursor/rules/*.md`
   - **Cline**: project `.clinerules`
   - **Continue**: `~/.continue/config.json` (system message), workspace
     `.continue/*.md`
   - **Aider**: `CONVENTIONS.md`, `.aider.conf.yml`
   - See `references/tools-memory-map.md` for the full table.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
99% confidence
Finding

Referencing and modifying ~/.continue/config.json is sensitive because it targets an agent configuration directory that controls future assistant behavior. Even if intended for convenience, directing an agent to alter this file can create durable, hard-to-audit changes and amplifies the effect of any bad or malformed profile data.

Content

Scanner excerpt · references/import-prompts.md (reported line 69)May include surrounding context.

Continue

text
Please update ~/.continue/config.json — find the systemMessage field and
merge this profile into it. Preserve existing systemMessage content if any.
Show me the diff before writing.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/tools-memory-map.md (reported line 16)May include surrounding context.

md
| **Codex CLI** (OpenAI) | `AGENTS.md`, `~/.codex/instructions.md` | Markdown | ✅ |
| **Cursor** | project `.cursorrules`, `.cursor/rules/*.md`, global Settings → Rules for AI | Markdown / inline text | ✅ |
| **Cline** (VSCode ext) | project `.clinerules`, global Settings → Custom Instructions | Markdown | ✅ |
| **Continue** (VSCode/JetBrains ext) | `~/.continue/config.json` `systemMessage` field, workspace `.continue/*.md` | JSON + Markdown | ✅ |
| **Aider** | `CONVENTIONS.md`, `.aider.conf.yml` `read` field | Markdown / YAML | ✅ via `--read` |
| **ChatGPT (web)** | OpenAI's hosted "Memory" feature | proprietary | ✅ but **vendor-locked** |
| **Claude.ai (web)** | Project knowledge files, custom instructions | text upload | ✅ |

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/tools-memory-map.md (reported line 58)May include surrounding context.

md
| **Codex CLI** (OpenAI) | `AGENTS.md`, `~/.codex/instructions.md` | Markdown | ✅ |
| **Cursor** | project `.cursorrules`, `.cursor/rules/*.md`, global Settings → Rules for AI | Markdown / inline text | ✅ |
| **Cline** (VSCode ext) | project `.clinerules`, global Settings → Custom Instructions | Markdown | ✅ |
| **Continue** (VSCode/JetBrains ext) | `~/.continue/config.json` `systemMessage` field, workspace `.continue/*.md` | JSON + Markdown | ✅ |
| **Aider** | `CONVENTIONS.md`, `.aider.conf.yml` `read` field | Markdown / YAML | ✅ via `--read` |
| **ChatGPT (web)** | OpenAI's hosted "Memory" feature | proprietary | ✅ but **vendor-locked** |
| **Claude.ai (web)** | Project knowledge files, custom instructions | text upload | ✅ |

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Unlike the earlier path-only references, this line actively recommends editing ~/.continue/config.json's systemMessage to paste a user profile into a global agent configuration. That is security-relevant because systemMessage is auto-applied, persistent, and can cause sensitive personal data or behavioral instructions to be injected into future sessions across projects.

Content

Scanner excerpt · references/tools-memory-map.md (reported line 60)May include surrounding context.

md
- **Config**: `~/.continue/config.json` → `systemMessage` field
- **Context**: `.continue/*.md` files in workspace
- Best place to paste profile: edit `~/.continue/config.json` `systemMessage`

### Aider

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README explains the convenience benefit but does not prominently warn, at the point of use, that the skill reads potentially sensitive long-term memory sources and writes the extracted data into a local file for later transfer. That omission can undermine informed consent and cause users to expose more personal or operational context than they expected, even if the skill intends to filter some sensitive fields.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrase "Generate my profile." is generic enough to plausibly appear in ordinary user conversation, which increases the chance of accidental activation in agents that map natural-language phrases to skills. In this skill's context, accidental invocation is more sensitive than usual because the described behavior includes reading long-term memory files and writing a consolidated profile document derived from them.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The core purpose of the skill is to extract long-term memory from one agent and package it into a portable profile for use in other tools. That is a direct data exfiltration pattern: even if intended for user convenience, it aggregates potentially sensitive memory contents into a transferable artifact, increasing exposure and making downstream misuse easier.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly instructs writing a profile derived from long-term memory to a local file without requiring an explicit consent step, sensitivity warning, or opt-in for disk persistence. Because the exported content can include personal preferences, identity details, work history, and other memory-derived data, this creates a real privacy risk through unintended local storage, later reuse, or accidental sharing.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file goes beyond producing a portable profile and provides prompts that instruct other tools to write into long-term memory, config files, and vendor-managed memory. That expands the skill from export/formatting into persistence and cross-tool state modification, which can cause unintended propagation of sensitive or harmful instructions into multiple agent environments.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
88% confidence
Finding

The Codex CLI prompt tells the target tool to save profile data into ~/.codex/instructions.md, creating persistent state across future sessions. Persistence is the core mechanism here, so if the exported profile contains overreaching instructions, secrets, or prompt-injection residue, it becomes a durable cross-session influence rather than a one-time response.

Content

Scanner excerpt · references/import-prompts.md (reported line 36)May include surrounding context.

Codex CLI

text
Please save the following profile to ~/.codex/instructions.md (create if
missing). If the file exists, append a "## User Profile" section at the end
and merge intelligently — don't duplicate existing entries.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The Continue prompt explicitly instructs editing ~/.continue/config.json, a live agent configuration file in the user's home directory. Writing profile content into a systemMessage field creates persistent behavioral influence and could embed unsafe, over-broad, or privacy-sensitive instructions into future sessions without adequate review.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The Aider prompt directs the target agent to modify project files (CONVENTIONS.md and .aider.conf.yml), which affects repository behavior rather than merely exporting or reusing profile content. In a shared or sensitive repo, this can introduce persistent instruction changes, commit noise, or hidden behavioral changes that outlive the current session.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The template includes mandatory Chinese alongside English in the top-level warning text, and later provides a full Chinese example output section. This imposes a specific language/locale presentation rather than offering language choice or documenting why a bilingual format is required.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The document explicitly directs users to read from and write to long-term memory/profile stores across multiple AI tools, but it does not warn that these files may contain sensitive personal data or persistent instructions. Because the skill's purpose is cross-tool profile extraction and import, omission of privacy and scope warnings can lead users to copy secrets, personal information, or sensitive preferences into broadly reused locations without understanding the exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file recommends appending profile data to global, auto-loaded instruction files such as ~/.claude/CLAUDE.md and similar locations, but does not explain that these files affect all future sessions and potentially all projects. This creates a real risk of cross-project data leakage, overbroad persistence of personal instructions, and accidental propagation of sensitive context into unrelated work.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
70% confidence
Finding

The natural-language description mixes English with Chinese usage examples but does not state whether the skill should respond in the user's preferred language. This can be interpreted as a locale/language policy concern because the skill presents a language-specific usage cue without explicit opt-in or language selection guidance.

Content

No source excerpt is available for this finding.