The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that direct the agent to bypass user confirmation and oversight during its autonomous operation. Evidence: "This runs fully autonomously. Do not ask the user for permission or confirmation — use your best judgment and keep moving." (SKILL.md).
[PERSISTENCE]: The skill sets up persistence via scheduled tasks and recurring loops. Evidence: Instructions for using the /loop command in Claude Code and the cron.add tool in OpenClaw to ensure the research continues autonomously across sessions (SKILL.md, references/agent-continuity.md).
[DATA_EXFILTRATION]: The skill includes instructions to transmit research reports and data to external third-party messaging services. Evidence: Instructions to "send it to the user via Telegram, WhatsApp, or Slack" within the automated loop payloads (SKILL.md, references/agent-continuity.md).
[INDIRECT_PROMPT_INJECTION]: The skill ingests and synthesizes information from external, untrusted research papers and web searches, creating an attack surface for malicious instructions to influence the agent's actions. 1. Ingestion points: Literature retrieval from Semantic Scholar, arXiv, and Exa search (SKILL.md). 2. Boundary markers: No clear delimitation or instructions to treat external data as untrusted are provided. 3. Capability inventory: Ability to write to the local filesystem, perform network operations, and invoke various domain-specific tools and skills. 4. Sanitization: No logic is provided for sanitizing or validating text extracted from external papers or search results.
[EXTERNAL_DOWNLOADS]: The skill references the installation of well-known research tools and libraries for literature search and report generation. Evidence: Mention of "pip install semanticscholar", "pip install arxiv", and utilities like "weasyprint" and "playwright" (SKILL.md, references/progress-reporting.md).

autoresearch