The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a vulnerability surface by ingesting untrusted data from the research-lookup tool and interpolating it directly into LaTeX documents for compilation.
Ingestion points: Data gathered from external sources via research-lookup.py (e.g., market trends, competitor financials, and regulatory data) is integrated into the final report body.
Boundary markers: Absent; the skill lacks specific delimiters or system instructions to ignore embedded malicious prompts within the researched data.
Capability inventory: The skill utilizes Bash and Write tools to execute xelatex, bibtex, and Python scripts, creating a path for code execution if a payload is successfully injected into the LaTeX source.
Sanitization: There is no evidence of filtering or sanitization performed on external data before it is written to the .tex files.
[DYNAMIC_EXECUTION]: The orchestration script scripts/generate_market_visuals.py uses subprocess.run() to dynamically execute Python scripts located in sibling directories (scientific-schematics and generate-image). While the script safely uses argument lists to mitigate shell injection, the reliance on dynamic path construction and sub-process spawning based on user-supplied topic strings introduces a standard automation risk.
[METADATA_POISONING]: There is a minor inconsistency between the declared author in the YAML frontmatter (K-Dense Inc.) and the platform-provided author context (marielynneblock). While likely thematic, such discrepancies can be used to mask the true origin of a skill.

market-research-reports