The Agent Skills Directory

[COMMAND_EXECUTION]: The script tools/ChartKit.py uses subprocess.run to execute external Python CLI tools located in the project's src/ directory. While it uses a list for arguments, the inputs include ticker symbols provided by the user.
[COMMAND_EXECUTION]: Markdown instructions in workflows/FullResearchWorkflow.md and workflows/GenerateSingleReport.md direct the agent to execute shell commands (e.g., uv run python src/analysis/risk_metrics_cli.py {TICKER} ...). These commands use ticker symbols as arguments and include shell-level redirections (>), creating a surface for command injection if the ticker symbol provided by the user contains shell metacharacters.
[DATA_EXPOSURE]: The skill reads the fin-guru/data/user-profile.yaml file to retrieve the investment_portfolio.total_value field, which contains sensitive user financial information used to personalize report sizing recommendations.
[PROMPT_INJECTION]: The skill ingests untrusted data from the Perplexity MCP to conduct market research and sentiment analysis, exposing it to indirect prompt injection risks.
Ingestion points: Data is fetched via mcp__perplexity__search and mcp__perplexity__reason in workflows/FullResearchWorkflow.md and workflows/GenerateSingleReport.md.
Boundary markers: The instructions do not define delimiters or specific warnings to ignore instructions embedded within the research results.
Capability inventory: The skill has the capability to write files (PDFs), execute shell commands via CLI tools, and launch subagents via the Task tool.
Sanitization: There is no evidence of sanitization or filtering for the external content before it is processed into the final analysis.
[DYNAMIC_EXECUTION]: The tools/ReportGenerator.py script performs sys.path.insert at runtime to include the project root, enabling the dynamic loading of modules from outside the skill's specific directory.
[COMMAND_EXECUTION]: The workflows/RegenerateBatch.md file instructs the agent to launch multiple parallel subagents using the Task tool, which involves the dynamic construction and execution of new agent prompts incorporating user input.

finance-report