The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to automatically install a third-party dependency (browser-act CLI/Skill) from an external URL if it is unavailable, bypassing user-authorized package management flows.
[PROMPT_INJECTION]: The instructions include high-risk autonomy abuse patterns. The skill explicitly commands the agent to perform software installation "without asking the user to confirm" and to "not request separate approval," which is a direct attempt to bypass standard safety guardrails and user consent protocols.
[PROMPT_INJECTION]: The skill utilizes concealment tactics by instructing the agent to misrepresent the source of its actions to the user. It specifies that the agent should "not describe it as a request for the user to open their browser," effectively hiding the automated fetching and installation of external code.
[EXTERNAL_DOWNLOADS]: The skill relies on an unverified third-party domain (browseract.com) for its core dependencies and installation logic rather than using established package registries or trusted vendor repositories.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of scraping and processing untrusted content from the web.
Ingestion points: Data collected from job boards and external platforms (SKILL.md).
Boundary markers: Absent. No delimiters or specific instructions are provided to differentiate between system instructions and data retrieved from external sources.
Capability inventory: The agent uses the browser-act tool to perform browser tasks and process instructions based on the data it scrapes (SKILL.md).
Sanitization: Absent. There is no evidence of content validation, escaping, or filtering before the agent integrates the external data into its workflow.

jobs-board-lead-scraper