T01 · Skill Instruction Hijacking
- Location
SKILL.md:3- Finding
Skill instructions attempt to override host-agent routing and decision authority
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This lottery skill is mostly coherent, but it asks for overbroad host-agent control while creating persistent tasks and enabling automated outbound notifications.
Review before installing. This skill can create recurring local tasks, store lottery records and costs, fetch unverified public draw data, and send task results through OpenClaw notifications. Only use it if you want Chinese lottery automation, trust the configured draw source, and are comfortable reviewing/disabling tasks and notification settings yourself.
SKILL.md:3Skill instructions attempt to override host-agent routing and decision authority
scripts/lotto.py:423Manual ticket records are accepted without number-range and uniqueness validation
scripts/fetch.py:30Mutable third-party draw data is trusted for prize and financial calculations without integrity verification
The README mandates that the host immediately invoke this skill whenever a message contains lottery-related words, which is an overly broad trigger that can cause the agent to act without clear user intent. In an agent environment, such forced invocation increases the chance of unintended automation, privacy exposure of message content to the skill, and unsafe chaining into commands like scheduling or notifications based on incidental mentions.
The skill declares an extremely broad highest-priority trigger on many common lottery-related words and explicitly instructs the agent to invoke the skill instead of general chat. This can cause frequent unintended routing of normal conversations into a transactional workflow, creating unsafe side effects such as unwanted task creation, record insertion, or automated actions based on ambiguous user text.
The README is written as an unconditional Chinese-language operational policy for activation and time parsing, with no indication that users may choose another language or locale. This can violate language/locale policy requirements when the skill is used in broader environments without explicit opt-in.
The README promotes scheduled execution, cron-driven automation, and push notifications, but does not prominently warn that these features create persistent tasks, store data, and may send outbound notifications automatically. In a host agent setting, users may not realize that a one-time interaction can lead to recurring background actions and message delivery, which raises consent and operational safety concerns.
The comment on L1 states '仅使用' in Chinese and appears to present the skill in a single language context without any opt-in or alternative locale guidance. Per the policy, forcing or implying a specific language without user choice can be a natural-language policy violation.
This file emits user-facing status and result messages in Chinese string literals, such as the database status text, without any visible language selection or opt-in mechanism. That creates a natural-language policy concern because the skill appears to force a specific locale on all users.
The cancel command invokes a destructive state-changing operation via store.cancel_recent and only reports the result after the action completes. In this file there is no pre-action confirmation prompt, warning message, or disclosure that recent records will be removed from cost and prize accounting.
The cancellation response message is hard-coded in Chinese and there is no visible mechanism in this file for users to choose another language. This indicates the CLI enforces a specific locale rather than offering or documenting a language preference.
The module docstring is written to indicate a China-specific lottery-fetching workflow and the file consistently uses Chinese-language natural-language strings for user-facing messages. There is no indication that users can choose another language or that the locale restriction is explicitly documented as a justified policy constraint.
The function fetches data from a remote URL via _read_json and then writes the resulting draw record to local storage with store.upsert_draw, but there is no confirmation prompt, warning-oriented log, or explanatory comment/docstring disclosing these external network and file/database side effects. Because this file is code, safety-relevant operations that transmit or persist data should include some visible disclosure unless clearly communicated elsewhere in code-facing text.
The file-level docstring is entirely in Chinese, and user-facing messages throughout the file are also Chinese-only. This forces a specific language/locale experience without any opt-in or indication that the skill is intentionally restricted to a Chinese-speaking or region-specific context.
The generate flow creates a batch UUID and inserts each ticket via store.insert_ticket, which writes user betting selections to persistent storage. There is no confirmation prompt, warning string, or explanatory comment here disclosing that the user's data will be saved, beyond a post-action follow-up that the numbers have already been recorded.
The record_manual flow parses user input and persists each ticket using store.insert_ticket. The only disclosure appears after the save in followup_messages, so users are not warned in advance that their manually entered betting data will be recorded.
The module docstring and all user-facing clarification messages are written only in Chinese, which implies the skill forces a specific language for interaction. Under the policy, locale or language constraints should be optional, user-selected, or clearly justified as region-specific.
This code file contains natural-language strings entirely in Chinese, beginning with the module docstring at L001, and there is no indication anywhere in the file that users may choose another language or that the skill is intentionally limited to a Chinese-only regional context. Under the policy, forcing a specific language without opt-in is a natural-language policy violation.
The skill includes an outbound messaging feature that can send arbitrary task-generated content to externally configured recipients through an external CLI. In this context, the capability is intentional, but it still creates a data-exfiltration and unauthorized-notification channel if an attacker can alter configuration, create tasks, or influence message content, especially because scheduled jobs can trigger pushes automatically.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
if account:
args.extend(["--account", account])
try:
proc = subprocess.run(args, text=True, capture_output=True, check=False)
except OSError as exc:
return {"ok": False, "sent": False, "error": str(exc)}
return {"ok": proc.returncode == 0, "sent": proc.returncode == 0,
The file hard-codes a China timezone (CN_TZ = timezone(timedelta(hours=8))) and uses it for all generated timestamps via now_iso() and cn_now(). This is a natural-language/policy concern because it imposes a specific locale setting globally without any visible user choice or justification in this file.
This JSON file contains numerous natural-language values such as names, labels, prize levels, and aliases exclusively in Chinese, including entries like "双色球", "红球", and "一等奖". Because the file provides no documented locale selection or opt-in mechanism, it may enforce a specific language/locale contrary to the policy for user choice.
The check_prize function calls store.update_ticket_prize to modify persisted ticket state and prize amounts. Although the final return message reports that prize checking completed, the code provides no explicit warning or confirmation that existing records will be updated as part of the operation.