T01 · Skill Instruction Hijacking
- Location
agent.py:399- Finding
Untrusted RSS Headlines Can Influence Downstream Agent Instructions
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent commodity-market analysis skill, but it relies on third-party data feeds and has some install and robustness risks users should understand.
Install only if you are comfortable with the skill contacting Yahoo Finance, Stooq, metals.live, Google News, and multiple RSS providers for market data. Prefer running it in a virtual environment, pinning or removing unused dependencies, and treating returned headlines as untrusted news text rather than instructions.
agent.py:399Untrusted RSS Headlines Can Influence Downstream Agent Instructions
agent.py:132Unbounded HTTP Response Reads Permit Memory and CPU Exhaustion
requirements.txt:1Unpinned and Unused Runtime Dependencies Increase Supply-Chain Exposure
The trigger includes the generic term "期货" alongside broad category terms like "大宗商品" and "commodity", which can appear in many everyday finance or news queries not specifically requesting this skill. The file does not provide exclusion conditions or narrower context boundaries, increasing the risk of unintended invocation.
Natural-language strings throughout the file, including the module description and generated output, are presented only in Chinese with no user language selection or opt-in. This can violate language/locale policy because the skill imposes a specific locale rather than offering a choice or documenting a justified regional scope.
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
{"name": "FXStreet 大宗商品", "url": "https://www.fxstreet.com/rss/news/commodity", "agency": "FXStreet", "country": "ES", "symbols_hint": ["XAUUSD","XAGUSD","CRUDEOIL"]},
{"name": "Rigzone 石油天然气", "url": "https://www.rigzone.com/news/rss/rigzone_latest.aspx", "agency": "Rigzone", "country": "US", "symbols_hint": ["CRUDEOIL","NATGAS"]},
{"name": "MarketWatch 市场", "url": "https://feeds.marketwatch.com/marketwatch/marketpulse/", "agency": "MarketWatch", "country": "US", "symbols_hint": ["XAUUSD","XAGUSD","CRUDEOIL","NATGAS","COPPER"]},
{"name": "MetalMiner 金属", "url": "https://agmetalminer.com/feed/", "agency": "MetalMiner", "country": "US", "symbols_hint": ["COPPER","XAGUSD"]},
]
NEWS_MAX_AGE = 7 * 24 * 3600
The skill performs numerous outbound requests to third-party market-data and news providers without explicit user disclosure or a consent/control mechanism. In an agent setting, this can leak usage patterns, IP/address metadata, timestamps, and potentially user-driven query intent to external services, increasing privacy and supply-chain exposure.
This function transmits requests to an external pricing API, introducing dependency on an unaffiliated remote service for runtime behavior and data integrity. In this skill's context, the request does not appear to include secrets, but it still exposes environment metadata and allows a third-party service to influence financial analysis output if the upstream data is wrong, manipulated, or unavailable.
def fetch_price_metals_live(cfg):
if cfg["yf"] not in ("GC=F", "SI=F"):
raise ValueError("metals.live: only gold/silver")
data = json.loads(http_get("https://api.metals.live/v1/spot", timeout=8))
key = "gold" if cfg["yf"] == "GC=F" else "silver"
price = data.get(key)
if price is None:
The trigger phrase "天然气" is very broad and can easily appear in ordinary user conversation about commodities, news, or general discussion. In an agent skill, such an ambiguous trigger can cause unintended invocation of the skill, leading to unexpected actions, confusing responses, or accidental use of external data sources when the user did not explicitly request the skill.
The trigger phrase "铜价" is also broad and likely to collide with normal market commentary or casual discussion about copper prices. Because this skill performs analysis functions, accidental triggering could cause the agent to route benign conversation into the skill workflow without a deliberate user request.
The natural-language description is presented exclusively in Chinese, and the file does not indicate that users may choose another language or locale. This can violate a language-choice policy where skills should not force a specific language without user opt-in.
The top-level docstring names the skill 'GoldSkill' and describes it as an 'international futures quantitative analysis system', which implies a gold-focused or at least more narrowly scoped tool. In practice, the code explicitly supports gold, silver, crude oil, natural gas, and copper, and also fetches general commodity news across many sources.
The dependency is specified with a lower-bound version only, which permits installation of any newer release and prevents reproducible builds. This increases supply-chain risk because a future breaking or compromised release could be pulled into the environment without review.
yfinance>=0.2.36
requests>=2.28.0
The requests package is also unpinned, so installations are not deterministic and may resolve to different versions over time. Because requests has a significant history of security advisories, leaving it unpinned raises the chance of deploying a vulnerable or otherwise unreviewed release.
yfinance>=0.2.36
requests>=2.28.0
The manifest includes requests without an exact version, while multiple known advisories exist for that package. Because the resolved installed version cannot be verified from this manifest, there is a real risk that deployments may pick an affected release, especially across different environments or rebuilds.