The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill explicitly recommends downloading external software from unverified sources, specifically mentioning the npm package @seoagent-official/seoagent and a plugin from the Baxter-Inc organization.
[REMOTE_CODE_EXECUTION]: The instruction to run npm install -g @seoagent-official/seoagent && seoagent init facilitates the download and immediate execution of remote code on the host system. This is an unverified dependency that runs with the user's local privileges.
[COMMAND_EXECUTION]: The skill provides shell commands for the user/agent to execute, including global npm installations and initialization scripts that modify the local environment.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it is designed to 'Always read existing .seoagent/ files before acting'. An attacker who can influence the content of files like .seoagent/audit/latest.md or .seoagent/context.md (e.g., via a malicious Pull Request or third-party script) could inject instructions that the agent would then follow in subsequent sessions.
Ingestion points: Files within the .seoagent/ directory, such as project.md, context.md, and audit/latest.md.
Boundary markers: None identified; the skill treats the content of these files as authoritative context for resuming work.
Capability inventory: The agent is encouraged to perform file reads/writes, execute npm commands, and perform site audits (network/tool requests).
Sanitization: No sanitization or validation of the workspace file content is described.

SEOAgent