The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from several sources, including user-provided research briefs (RESEARCH_BRIEF.md), command-line arguments, and external research paper content/metadata fetched from arXiv. This data is used as primary context for automated reasoning phases, which could allow maliciously crafted content within a research paper or brief to influence agent behavior.
Ingestion points: Processes content from RESEARCH_BRIEF.md, $ARGUMENTS, and external data fetched via the /arxiv and /research-lit tools.
Boundary markers: The workflow uses Markdown sections and headers to organize data, but it does not employ explicit delimiters or instructions to ignore potential commands embedded within the external research content.
Capability inventory: The skill has the ability to execute subprocesses (via python3), perform network requests to research repositories, and write files to the project directory.
Sanitization: There is no evidence of sanitization or filtering of external content before it is interpolated into prompts for sub-skills.
[COMMAND_EXECUTION]: The skill invokes local Python scripts (run_state.py, idea_discovery_gate.py, reviewer_route.py) via python3 to handle state persistence and evidence validation. It also utilizes shell commands (cat << 'EOF') as a fallback mechanism for writing large files to the filesystem.

idea-discovery