The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves using the research-lookup tool to ingest external literature and data, which is then processed by the agent to generate manuscripts and write files. Because this data originates from untrusted external sources and the skill possesses write_file capabilities, it presents a potential surface for indirect prompt injection attacks where malicious instructions hidden in research papers could attempt to influence agent actions.
Ingestion points: External research data gathered via research-lookup (referenced in SKILL.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands in the ingested data were found in the provided files.
Capability inventory: The skill uses read_file, search_project, write_file, and references the execution of a Python script via shell.
Sanitization: No sanitization or validation protocols for the external data are mentioned.
[COMMAND_EXECUTION]: The instructions include a mandatory requirement for the agent to execute a shell command to generate figures using an internal script.
Evidence: SKILL.md provides the command: python scripts/generate_schematic.py "your diagram description" -o figures/output.png.
Metadata Discrepancy: The YAML frontmatter in SKILL.md states scriptCount: 0, which contradicts the instructions referencing a script located at scripts/generate_schematic.py. While likely a configuration error, this represents a lack of transparency in the skill's defined capabilities.

scientific-writing