The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the research-lookup tool and processes it into the agent's context and final report outputs.
Ingestion points: Data enters the agent context via outputs from research-lookup (SKILL.md, Phase 1).
Boundary markers: No specific delimiters or "ignore instructions" warnings are used in the prompts that handle external research data.
Capability inventory: The skill utilizes Bash (for script execution), Write, and Edit (for creating LaTeX files).
Sanitization: There is no evidence of sanitization or validation for the data retrieved from external sources before it is used to generate the report.
[COMMAND_EXECUTION]: The skill uses Bash to execute multiple Python scripts with arguments derived from user input or research data. For example, python ... --topic "[MARKET NAME]" and python ... "[IMAGE PROMPT]". This creates a surface for command injection if the interpolated strings contain shell metacharacters (e.g., semicolons, backticks).
[REMOTE_CODE_EXECUTION]: The skill depends on and executes code from other external skill directories, such as skills/scientific-schematics/scripts/, skills/generate-image/scripts/, and skills/research-lookup/scripts/. This execution of code outside the skill's own directory requires trust in those external skill modules.
[DYNAMIC_EXECUTION]: The skill follows a pattern of dynamic script generation and execution by writing LaTeX (.tex) files based on research data and then executing the xelatex compiler and bibtex to generate a PDF. This involves the execution of document code that incorporates untrusted external content.

market-research-reports