The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured JSON data from a remote API at market.ft.tech. This creates a potential surface for indirect prompt injection if the remote data source were compromised to return malicious instructions.
Ingestion points: Data enters the agent context through scripts/handler.py across all sub-skills.
Boundary markers: Absent; the skill does not explicitly wrap or tag external content as untrusted.
Capability inventory: Use of runpy.run_path for execution and urllib.request.urlopen for network access.
Sanitization: The skill decodes JSON but does not filter for natural language instructions in data fields.
[DYNAMIC_EXECUTION]: The entry point run.py utilizes runpy.run_path to dynamically load and execute handler scripts based on user-supplied sub-skill names. This behavior is mitigated by an explicit whitelist check that restricts execution to files found in the sub-skills/ directory.
[DATA_EXFILTRATION]: The skill performs HTTP GET requests to external endpoints under the market.ft.tech domain. These operations are essential for its primary function of retrieving market data and are restricted by a safe_urlopen wrapper that validates the target domain against the configured base URL.

ftshare-market-data