The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its core function of fetching and analyzing untrusted external data.
Ingestion points: The workflow relies on the WebSearch and WebFetch tools to retrieve company data from the internet and uses the $ARGUMENTS variable for user-supplied research targets.
Boundary markers: There are no instructions to use specific delimiters or "ignore embedded instructions" warnings when the agent or its sub-agents process content retrieved from external websites.
Capability inventory: The skill possesses significant capabilities, including writing reports to the filesystem (Write), executing shell commands via Bash for local repository tools, and spawning autonomous sub-agents (Agent, TaskCreate).
Sanitization: The instructions do not mandate any filtering or sanitization of external content before it is interpolated into prompts or used for decision-making.
[COMMAND_EXECUTION]: The skill instructions direct the agent to utilize shell commands to interact with the local environment and execute specialized tools.
Evidence: The "Codex adapter note" and research steps explicitly instruct the agent to run the date command and execute local scripts such as python3 tools/financial_rigor.py from the repository root.
Context: These executions target internal vendor-provided scripts (tools/) and standard system utilities to perform the skill's primary research tasks.

private-company-research