The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions in SKILL.md and references/financial.md authorize the agent to write and execute "bespoke modules" or "code written on the fly" to handle financial modeling scenarios that exceed the capabilities of the provided script. This runtime script generation and execution can be risky if the logic is influenced by data from untrusted user artifacts.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core workflow of ingesting and analyzing untrusted external data.
Ingestion points: Artifacts gathered in references/intake.md including resumes, performance reviews, portfolios, and financial records.
Boundary markers: SKILL.md and references/analysis-agents.md require agents to cite specific evidence tags, which acts as a check against accidental obedience to instructions embedded within artifacts.
Capability inventory: The skill uses file I/O to a workspace directory, performs web searches for financial data, spawns autonomous sub-agents, and executes Python code (including dynamically generated modules).
Sanitization: There is no evidence of sanitization or filtering applied to the user-provided artifacts before they are read by the analysis agents.
[DATA_EXFİLTRATION]: The skill systematically collects highly sensitive data including net worth statements, income records, and detailed performance reviews, storing them in a local workspace directory (./life-paths-workspace/). While the data remains local by default, the concentration of high-value PII and financial information significantly increases the impact of any potential exfiltration via tool misuse or injection.
[COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/fi_model.py) and utilizes python3 -m unittest for automated testing of both bundled and dynamically generated code.

life-paths