The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements persistence and continuity mechanisms through platform-specific scheduling commands.\n
Instructions in SKILL.md and references/agent-continuity.md direct the agent to initialize a /loop in Claude Code and a 20-minute cron.add job in OpenClaw to ensure research tasks continue autonomously.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python libraries to facilitate research data retrieval and report generation.\n
SKILL.md and references/skill-routing.md instruct the agent to install packages such as semanticscholar, arxiv, and PDF generation tools like weasyprint.\n- [PROMPT_INJECTION]: The skill contains instructions that promote high autonomy and the removal of standard human-in-the-loop interaction constraints.\n
SKILL.md includes the directive: "Do not ask the user for permission or confirmation — use your best judgment and keep moving. The human is asleep or busy; your job is to make as much research progress as possible on your own."\n- [DATA_EXFILTRATION]: The skill facilitates the transmission of project data and reports to external third-party communication platforms.\n
SKILL.md and references/agent-continuity.md instruct the agent to send research updates, plots, and PDF reports to the user via Telegram, WhatsApp, or Slack, involving network transmission to non-whitelisted domains.\n- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its automated processing of external research content.\n
Ingestion points: Research papers and metadata are retrieved from the web via web_search_exa, Semantic Scholar, and arXiv APIs and stored in the literature/ directory.\n
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions that may be embedded in the ingested research papers.\n
Capability inventory: The agent has access to shell execution (/loop, cron.add), package installation, and routing to specialized domain skills (infrastructure, training, etc.).\n
Sanitization: No explicit sanitization or filtering is performed on the content retrieved from external research sources before integration into the agent's context.

autoresearch