The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses a sensitive local directory at ~/.easel-browser-profiles/ZhihuProfile. This directory contains the persistent browser profile, including session cookies and authentication data for the Zhihu platform. Accessing this path allows for the retrieval of credentials necessary for automated publishing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the Zhihu platform, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Scrapes page text and metadata using page.inner_text("body") and locators defined in SKILL.md and references/commands.md.
Boundary markers: No explicit delimiters or instructions are used to distinguish between user commands and data ingested from the web.
Capability inventory: The skill can automate browser interactions, write files to the outputs/ directory, and execute local Python scripts (zhihu_answer.py, persona_gate.py, log.py).
Sanitization: The ingested web content is processed without filtering or validation to strip potentially malicious strings.
[DYNAMIC_EXECUTION]: Uses page.evaluate() to run Javascript strings in the browser context. This is used to bypass UI obstructions (headers) and find buttons based on their text labels.
[COMMAND_EXECUTION]: Launches Chromium via Playwright with the --no-sandbox flag, which reduces the security isolation of the browser process.

skill-zhihu-answer