The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a persistent execution mechanism using the /loop command (Claude Code) and cron.add (OpenClaw). These instructions require the agent to run automated research tasks at 20-minute intervals continuously, maintaining its own execution state across sessions.
[PROMPT_INJECTION]: Instructions explicitly direct the agent to operate autonomously and bypass user confirmation ("Do not ask the user for permission or confirmation"). This removes human oversight from potentially sensitive actions such as file system modifications or data transmission.
[DATA_EXFILTRATION]: The automated reporting workflow includes instructions to transmit research progress and summaries to external messaging platforms (Telegram, WhatsApp, Slack), involving network operations to non-whitelisted domains.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python packages (semanticscholar, arxiv) and system utilities (weasyprint, playwright) to support paper retrieval and report generation. These downloads target well-known academic and technology services.
[PROMPT_INJECTION]: The skill ingests and processes summaries of external research papers from the web to guide its decision-making and hypothesis formation, creating a surface for indirect prompt injection.
Ingestion points: Summaries of external papers stored in the literature/ directory.
Boundary markers: None; ingested content is integrated into the research narrative without delimiters or instructions to ignore embedded commands.
Capability inventory: The agent has the ability to write to the local file system, install software packages, and communicate with external messaging APIs.
Sanitization: No validation or filtering is performed on external paper data before it is processed and used to determine research direction.

autoresearch