ClawHub
Back to skill

Security audit

Fetch & Archive to Lexiang

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent archiving purpose, but it uses broad browser-session and credential access with weak controls that users should review carefully before installing.

Install only if you intentionally want an archiving tool that can use your logged-in browser and Lexiang credentials. Prefer running it in an isolated browser profile, avoid using your daily Chrome session, review any ~/.fetch_article helper before use, and do not use it where WorkBuddy, Cursor, Lexiang, or browser credentials could expose sensitive organizational data.

A.I.G

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Findings (8)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
references/lessons-learned.md:293
Finding

Live Lexiang access-token harvesting from WorkBuddy application logs

Content
View full analysis
/workbuddyMainThread__*.log by finding the latest line containing X-Oneid-Access-Token. Self-check: Always obtain the latest token from the logs. ``` ### Technical Analysis These instructions tell the Agent to search another application's logs for a live bearer token and use it for direct HTTP requests. This bypasses supported credential delegation and authentication interfaces. The behavior violates least privilege because archival does not require exposing the raw token to the Agent, its shell commands, or its conversational context. The same project provides a safer approach through delegated Lexiang tools and the authenticated uploader. Although the beginning of `lessons-learned.md` describes direct MCP access as historical, this section still calls log extraction the correct method and repeats it as a mandatory self-check. That contradiction may cause an Agent to follow the unsafe instruction. ### Attack Path 1. An attacker or user requests an archival operation in a WorkBuddy sandbox. 2. The Skill requires the Agent to read `references/lessons-learned.md`. 3. The Agent follows the instruction to search `~/.workbuddy/logs/`. 4. It extracts the latest `X-Oneid-Access-Token` value. 5. The raw bearer token enters Agent context, shell history, or HTTP request construction. 6. A malicious prompt, compromised tool, or observer can reuse the token to access Lexiang resources within the token's authorization scope. ### Impact Assessment Successful exploitation can expose an active Lexiang bearer token and permit A ...[truncated 242 chars]
Remediation
View remediation

T07 · Tool Hijacking and Spoofing

Error
Location
scripts/fetch_article.py:233
Finding

Automatic execution of an unaudited shell script from the user home directory

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/fetch_article.py:155
Finding

Complete Chrome cookie database copied into a persistent remote-debugging profile

Content
View full analysis
None: cdp_profile_dir = _cdp_profile_dir() real_chrome_dir = _get_chrome_user_data_dir() real_cookies = Path(real_chrome_dir) / "Default" / "Cookies" cdp_cookies = Path(cdp_profile_dir) / "Default" / "Cookies" if real_cookies.exists(): os.makedirs(str(cdp_cookies.parent), exist_ok=True) try: shutil.copy2(str(real_cookies), str(cdp_cookies)) except Exception as e: print(f"Cookie synchronization failed: {e}") def _launch_cdp_chrome_via_open(port: int = CDP_PORT) -> bool: cdp_profile_dir = _cdp_profile_dir() os.makedirs(cdp_profile_dir, exist_ok=True) _sync_cookies_to_cdp_profile() cmd = [ chrome_path, f"--remote-debugging-port={port}", f"--user-data-dir={cdp_profile_dir}", "--no-first-run", "--no-default-browser-check", "--disable-default-apps", ] global _chrome_process _chrome_process = subprocess.Popen( cmd, stdout=subprocess.DEVNULL, stderr=subprocess.PIPE, ) ``` Messages and comments have been omitted or normalized to English; the security-relevant executable code is unchanged. ### Technical Analysis The fetcher copies the entire cookie database from the user's daily Chrome profile into a persistent CDP profile. The operation is not restricted to the requested website or the minimum cookies needed for that source. The copied profile is then started with a remote-debugging port. A process capable of reaching that local debugging endpoint may control tabs, evaluate JavaScript, inspect browser storage, and interact with authenticated sessions available in the profile. This exceeds the minimum privilege necessary to fetch one authentica ...[truncated 988 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/upload_video_via_openapi.py:65
Finding

Unvalidated API base can receive Lexiang application secrets and bearer tokens

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/upload_video_via_openapi.py:262
Finding

Bearer access-token material is printed to terminal and Agent logs

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/upload_video_via_openapi.py:128
Finding

Token cache is written before restrictive permissions are applied

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/md_to_pdf.py:146
Finding

Markdown image path traversal permits embedding local files into generated PDFs

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
references/lessons-learned.md:3
Finding

User feedback can be converted into mandatory persistent Skill instructions

Content
View full analysis
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (73)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented workflow instructs the agent to reuse the user's real Chrome session via CDP, read a port file under the home directory, and rely on existing login state to fetch protected content, while the high-level description does not disclose those sensitive behaviors. Hidden access to authenticated browser context materially changes the trust model because it can expose cookies, account data, and privileged sessions beyond what a user would expect from a generic content-fetching skill.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/lessons-learned.md (reported line 137)May include surrounding context.

md
2. `~/.lexiang/openapi.json` 只需 `app_key` / `app_secret` / `staff_id`，`api_base` 有默认值 `https://lxapi.lexiangla.com`
  3. token 接口必须用 JSON body（已在脚本中修正）
  4. 上传后 VOD 转码需要几秒到几分钟，`status` 从 `processing` 变为 `finished` 后可播放
  5. 删除旧条目可用 OpenAPI `DELETE /cgi-bin/v1/kb/entries/{entry_id}?space_id=xxx`
- **已修复**：`upload_video_via_openapi.py` 的 `DEFAULT_API_BASE` 和 `get_access_token` 函数
- **同步更新**：本文件；SKILL.md youtube 视频流程说明

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · references/youtube-video.md (reported line 29)May include surrounding context.

md和meta.json`。

它不翻译、不生成最终双语稿、不上传 Markdown。旧 --skip-translate 仅兼容旧命令， 会提示 deprecated。

命令

bash
python3 scripts/yt_download_transcribe.py "<YouTube URL>" \
  --output-dir <work-dir> \
  --whisper-model base

常用参数：

  • --whisper-model tiny|base|small|medium|large
  • --skip-download
  • --keep-audio
  • --cookies-from-browser chrome

输出

text
<work-dir>/
├── source.md
├── meta.json
└── <视频标题>.mp4

source.md 结构为元信息、## 视频介绍、## 逐字稿。转录段落含时间戳，不插入翻译 emoji，也不在每段之间插入分隔线。

meta.json 包含标准字段：

  • title
  • source_url
  • source_title
  • source_type: youtube
  • language

以及频道、发布日期、时长、缩略图、媒体路径等扩展字段。

如果目标目录已有内容不同的 source.md，脚本报错，要求使用新目录。

内容

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill explicitly advertises bypassing paywalls, anti-automation detection, and bot protections by reusing a real browser's authenticated state. In this skill context, that is not neutral archival; it is purposeful circumvention of access controls and strongly elevates the abuse potential of the entire tool.

Content

No source excerpt is available for this finding.

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
99% confidence
Finding

The YARA hit is justified because the file contains multiple information-stealer patterns: reading Chrome cookies, obtaining Keychain material, decrypting stored cookies, syncing them into another profile, and reusing authenticated sessions. In context, these are not incidental strings; they correspond to implemented credential-harvesting functionality.

Content

Scanner excerpt · scripts/fetch_article.py (reported line 14)May include surrounding context.

python
ol (CDP)
   on port 9222. This completely bypasses Google's anti-automation detection and
   Cloudflare's bot detection because it IS the user's real Chrome — with full
   login state (cookies, localStorage, etc.).
   Requires Chrome to be running with --remote-debugging-port=9222.
   Automatically detects and waits for Cloudflare JS challenges.

2. **Cookie injection mode** (default):
   Reads cookies from Chrome's cookie database and injects them into a fresh
   Playwright browser context. Works for most sites but Google login pages
   and Cloudflare-protected sites may detect the automation and block it.

Image handling:
   Downloaded images are automatically inspected for format mismatches
   (e.g. WebP served as .png, SVG served as .png) and converted to real
   PNG for maximum compatibility with PDF generators and document editors.

Usage:
    # CDP mode (for OpenAI, LinkedIn, Cloudflare-protected sites)
    python fetch_article.py fetch <URL> --output-dir <dir> --cdp

    # Co

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code launches or connects to the user's real Chrome over remote debugging specifically to inherit cookies, localStorage, and login state. This gives the script effective control over an authenticated browser session and can access any site the user is logged into, substantially expanding blast radius.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This code copies Chrome's cookie database, derives decryption keys, decrypts cookies, and reuses them in automation. That is credential/session theft behavior in practice because it extracts bearer credentials from the user's browser and repurposes them outside the original browser context.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The tool accesses and decrypts browser cookies without an explicit, prominent warning at the point of use that it will read sensitive session credentials from the local browser profile. Even if intended for convenience, this is highly sensitive behavior that users may not reasonably expect from an article fetcher.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

This finding correctly identifies credential access behavior: the function is specifically built to retrieve browser cookies, which are authentication artifacts. In the context of bypassing paywalls and bot controls, this materially enables account/session abuse.

Content

Scanner excerpt · scripts/fetch_article.py (reported line 342)May include surrounding context.

python
def get_chrome_cookies_for_domain(domain: str) -> list[dict]:
    """Extract cookies for a domain from Chrome's cookie database.
    
    On macOS, Chrome encrypts cookies with the Keychain.
    We use a simpler approach: copy the Cookies db and read what we can,
    or use the `security` command to decrypt.
    """

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Reading Keychain material to decrypt browser cookies is a privileged credential-access action, yet the code performs it without clear disclosure or granular consent. That creates a significant privacy and account-compromise risk if users run the skill without understanding its behavior.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

Accessing Keychain to obtain the Chrome Safe Storage secret is a direct step in decrypting stored browser credentials. This elevates the code from simple scraping into active credential extraction and session hijacking capability.

Content

Scanner excerpt · scripts/fetch_article.py (reported line 378)May include surrounding context.

python
return []

        # On macOS, cookies are encrypted. We need to decrypt them.
        # Get the Chrome Safe Storage key from Keychain
        try:
            result = subprocess.run(
                [

MCP Config Access

High
Category
Agent Snooping
Confidence
98% confidence
Finding

Direct access to MCP config is a sensitive capability because it exposes reusable authentication material from a general-purpose local tool configuration. In this skill context, that means the script can act with the user's existing Lexiang privileges to enumerate entries, fetch parsed content, and download PDFs, making the credential access materially security-relevant rather than incidental.

Content

Scanner excerpt · scripts/lexiang_pdf_parse.py (reported line 29)May include surrounding context.

python
def load_token():
    """从 ~/.cursor/mcp.json 读取乐享 MCP token（去掉 Bearer 前缀）。"""
    p = os.path.expanduser("~/.cursor/mcp.json")
    d = json.load(open(p))
    auth = d["mcpServers"]["lexiang"]["headers"]["Authorization"]

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
89% confidence
Finding

The script defaults to extracting authentication cookies from the local Chrome browser and passing them to yt-dlp for both metadata retrieval and download operations. In an agent skill context, this is sensitive because it silently expands access from public content to any account-authenticated content available in the operator's browser session, creating a real risk of unauthorized data access or privacy leakage if the skill is run on a user workstation or shared automation host.

Content

Scanner excerpt · scripts/yt_download_transcribe.py (reported line 60)May include surrounding context.

python
# 检查 ffmpeg
    try:
        subprocess.run(["ffmpeg", "-version"], capture_output=True, check=True)
    except (subprocess.CalledProcessError, FileNotFoundError):
        missing.append("ffmpeg (brew install ffmpeg)")

    if missing:
        print("❌ 缺少以下依赖：")
        for dep in missing:
            print(f"   - {dep}")
        sys.exit(1)


def get_video_info(url: str, cookies_from_browser: str | None = "chrome") -> dict:
    """获取视频元信息（标题、时长、频道等）"""
    print("📋 获取视频信息...")
    cmd = ["yt-dlp", "--dump-json", "--no-download"]
    if cookies_from_browser:
        cmd.extend(["--cookies-from-browser", cookies_from_browser])
    cmd.append(url)
    result = subprocess.run(cmd, capture_output=True, text=True)
    if result.returncode != 0:
        # 如果带 cookies 失败，尝试不带 cookies
        if cookies_from_browser:
            print("   ⚠️  使用浏览器 cookies 失败，尝试不带 cookies

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README explicitly documents uploading archived content to Lexiang and relying on stored credentials, but it does not warn users that content and metadata may be transmitted to external services or published into a shared knowledge base. In a skill that handles articles, videos, podcasts, and PDFs, this omission increases the risk of users unintentionally sending sensitive, proprietary, or copyrighted material to a remote system.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill declares broad operational behavior involving shell execution, filesystem access, network fetching, browser/CDP control, and uploads, but it does not define any explicit tool scope such as permissions or allowed-tools. That creates an avoidable over-privilege condition where an agent may invoke sensitive capabilities without a clear least-privilege boundary, increasing the chance of misuse or unsafe execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill directs the agent to fetch remote content, download media, transcribe audio, upload documents, and reuse an authenticated Chrome context, but it does not include a clear user-facing warning about external data transfer, third-party processing, or the use of the user's logged-in browser session. In this context, the omission is more dangerous because the skill handles potentially sensitive browsing/account state and may transmit content to external services or internal platforms without explicit informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The content explicitly endorses minimizing human code review and relying on agents to decide when escalation is necessary. In a skill about agentic engineering and archival workflows, that guidance can normalize weaker review controls and increase the chance that insecure or incorrect code is accepted without meaningful human scrutiny.

Content

No source excerpt is available for this finding.

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · output/kun-chen-agentic-engineering/Agentic Engineering, Clearly Explained in 60 Min by an Ex-Meta L8 Engineer - Kun Chen.md (reported line 100)May include surrounding context.

md
<p>Dear subscribers,</p><p>Today, I want to share a new episode with Kun Chen.</p><p><a href="https://x.com/kunchenguid">Kun </a>is an ex-L8 principal engineer at Meta who now ships up to 40 PRs a day while rarely reviewing code. In our episode, he walked through the free tools he built to make that possible: <a href="https://github.com/kunchenguid/lavish-axi">Lavish</a> for visual planning in HTML, <a href="https://github.com/kunchenguid/treehouse">Treehouse</a> for parallel agents, and <a href="https://github.com/kunchenguid/no-mistakes">No Mistakes</a> for catching AI’s errors before they make it to production.</p><p>Watch now on <strong><a href="https://youtu.be/88B6DimMD2g">YouTube</a></strong>, <strong><a href="https://podcasts.apple.com/us/podcast/behind-the-craft/id1736359687?ign-itscg=30200&amp;ign-itsct=podtail_podcasts">Apple</a></strong>, and <strong><a href="https://open.spotify.com/episode/2GdrQsplBgFPLTyzK88Zy9?si=02WmUhzXTvCysGkuflTE9g">Spotify</a></strong>.</p><div id="youtube2-88B6DimMD2g" class="youtube-wrap" data-attrs="{&quot;videoId&quot;:&quot;88B6DimMD2g&quot;,&quot;startTime&quot;:null,&quot;endTime&quot;:null}" data-component-name="Youtube2ToDOM"><div class="youtube-inner"><iframe src="https://www.youtube-nocookie.com/embed/88B6DimMD2g?rel=0&amp;autoplay=0&amp;showinfo=0&amp;enablejsapi=0" frameborder="0" loading="lazy" gesture="media" allow="autoplay; fullscreen" allowautoplay="true" allowfullscreen="true" width="728" height="409"></iframe></div></div><p>Kun and I talked about:</p><ul><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g">00:00</a>) Why he doesn't review code anymore</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=64s">01:04</a>) Agentic engineering: Plan, code, validate</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=382s">06:22</a>) Demo: Fixing an AI tutor screen with agents</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=520s">08:40
...[truncated 27 chars]

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · output/kun-chen-agentic-engineering/body_full.html (reported line 1)May include surrounding context.

html
<p>Dear subscribers,</p><p>Today, I want to share a new episode with Kun Chen.</p><p><a href="https://x.com/kunchenguid">Kun </a>is an ex-L8 principal engineer at Meta who now ships up to 40 PRs a day while rarely reviewing code. In our episode, he walked through the free tools he built to make that possible: <a href="https://github.com/kunchenguid/lavish-axi">Lavish</a> for visual planning in HTML, <a href="https://github.com/kunchenguid/treehouse">Treehouse</a> for parallel agents, and <a href="https://github.com/kunchenguid/no-mistakes">No Mistakes</a> for catching AI’s errors before they make it to production.</p><p>Watch now on <strong><a href="https://youtu.be/88B6DimMD2g">YouTube</a></strong>, <strong><a href="https://podcasts.apple.com/us/podcast/behind-the-craft/id1736359687?ign-itscg=30200&amp;ign-itsct=podtail_podcasts">Apple</a></strong>, and <strong><a href="https://open.spotify.com/episode/2GdrQsplBgFPLTyzK88Zy9?si=02WmUhzXTvCysGkuflTE9g">Spotify</a></strong>.</p><div id="youtube2-88B6DimMD2g" class="youtube-wrap" data-attrs="{&quot;videoId&quot;:&quot;88B6DimMD2g&quot;,&quot;startTime&quot;:null,&quot;endTime&quot;:null}" data-component-name="Youtube2ToDOM"><div class="youtube-inner"><iframe src="https://www.youtube-nocookie.com/embed/88B6DimMD2g?rel=0&amp;autoplay=0&amp;showinfo=0&amp;enablejsapi=0" frameborder="0" loading="lazy" gesture="media" allow="autoplay; fullscreen" allowautoplay="true" allowfullscreen="true" width="728" height="409"></iframe></div></div><p>Kun and I talked about:</p><ul><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g">00:00</a>) Why he doesn't review code anymore</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=64s">01:04</a>) Agentic engineering: Plan, code, validate</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=382s">06:22</a>) Demo: Fixing an AI tutor screen with agents</p></li><li><p>(<a href="https://www.youtube.com/watch?v=88B6DimMD2g&amp;t=520s">08:40
...[truncated 27 chars]

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical
Code
suspicious.dynamic_code_execution
Location
tests/test_contracts.py:17