Security audit
Security checks for vulnerabilities and agentic risk
This Twitter timeline digest skill is not malicious, but it needs review because it uses an authenticated Twitter session and stores timeline-derived data locally while its trigger and disclosures are too broad.
Review before installing. Use this only if you are comfortable letting an agent invoke bird with your logged-in X/Twitter session, fetch personalized timeline content, and write state or JSON files locally. Keep invocation limited to timeline digest tasks, avoid shared machines for authenticated runs, and periodically inspect or delete ~/.timeline-digest plus any raw timeline JSON outputs.
The trigger condition says to use the skill for broad needs like data analysis, reporting, statistics, and visualization, which are far outside the documented Twitter timeline digest scope. Overbroad routing increases the chance an agent invokes this skill on unrelated or sensitive tasks, unnecessarily exposing authenticated timeline data and causing unexpected local execution.
L058-L069 明确把免费版描述为仅输出原始JSON、且“不支持”分类标注；但 L477-L489 又将 LLM API 标为“必需”，并说明可提供 LLM 智能摘要功能。这不是单纯信息缺失，而是对免费版是否包含智能摘要能力的直接冲突，容易误导使用者对技能实际能力与依赖的理解。
The skill relies on an authenticated bird CLI session using cookie-based login to access private or personalized X/Twitter timeline data, but it does not provide a prominent privacy warning. This can mislead users into exposing highly personalized account data to local tooling and downstream summarization flows without informed consent.
The skill instructs creation of local configuration and state directories but does not clearly warn users that execution will persist data on disk. Without prominent disclosure, authenticated timeline-derived metadata may accumulate locally and remain accessible to other local processes or users.
The skill explicitly creates a persistent directory and maintains state across runs, enabling session-like retention of processed tweet IDs and timestamps. In the context of authenticated social-media timeline processing, this persistence increases privacy risk, leaves forensic residue on disk, and may expose behavioral history if the local machine or agent workspace is shared or compromised.
} EOF
mkdir -p ~/.timeline-digest
### 首次运行
The FAQ explicitly frames output as generating Chinese classified summaries, while the broader document is written as if Chinese is the required interaction/output language and does not offer user language selection. This can violate language/locale policy when a skill forces a specific language without opt-in.
The skill claims to be 'read-only' while the examples and logic clearly create and update persistent local state files. This misrepresentation can cause users or agents to authorize execution under false assumptions, leading to unintended data retention and writes on the host filesystem.