The Agent Skills Directory

[SAFE]: The skill follows secure practices by using Python scripts to handle research logic locally. All file system operations are restricted to paths provided via command-line arguments, and the skill does not perform its own network requests, mitigating data exfiltration risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (excerpts and titles from research sources) provided by the host agent. This content is written to a persistent JSON state file and eventually rendered into a markdown report. While this creates a potential surface for indirect prompt injection if the agent processes the generated report as a set of instructions, the skill treats this data as static text strings.
Ingestion points: scripts/iterate.py accepts JSON input via the --record flag which contains source titles and excerpts.
Boundary markers: The Markdown report produced by scripts/compile.py does not use explicit boundary markers to delimit untrusted source content.
Capability inventory: The skill uses Python scripts to automate state management and report generation but does not invoke external shell commands or network utilities directly.
Sanitization: The scripts do not perform explicit sanitization or instruction filtering on the research excerpts before rendering them into the final report.

deep-research