The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves analyzing untrusted third-party data from competitor websites and technical articles.
Ingestion points: The agent is instructed to perform 'runtime forensics' including structure extraction (DOM), resource supply analysis, and response probing on external websites.
Boundary markers: The instructions do not specify technical delimiters or strict 'ignore' instructions for the data ingested from these external sources.
Capability inventory: The skill leverages external tools for web browsing and network requests (via the effective-web-research dependency).
Sanitization: While the methodology emphasizes cross-validation and first-hand evidence to ensure factual accuracy, there are no specific technical filters to prevent an attacker-controlled website from embedding instructions (e.g., in HTML comments or metadata) to manipulate the agent's summary or next steps.
[EXTERNAL_DOWNLOADS]: The skill identifies a prerequisite dependency, effective-web-research, and instructs the agent to print an installation command (npx skills add FuDesign2008/open-skills -g) if the dependency is missing. This command downloads code from an external repository managed by the same author.
[COMMAND_EXECUTION]: The methodology describes several technical verification steps, such as 'runtime isolation probes' and 'resource supply analysis' using browser devtools or equivalent automation. While these are presented as research tasks for the agent to perform, they involve interacting with potentially malicious environments (competitor sites) through the browser.

competitive-research