The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions require the execution of shell scripts located in the environment at workspace/scripts/code-debugger-task-bootstrap.sh and workspace/scripts/deliver-report.sh. These scripts are not included in the skill package, and their logic cannot be verified during static analysis.
[DATA_EXFILTRATION]: The skill includes a dedicated phase for 'Feishu Delivery', which involves copying generated forecast reports and sending them to an external communication platform (Feishu/Lark). While this is the intended functionality, it constitutes a data exfiltration vector for the analyzed intelligence.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its multi-round intelligence gathering process.
Ingestion points: Phase 1 involves multiple rounds of keyword searches across various web providers (Brave, Tavily, Zhipu, etc.), with raw results stored in sources/raw/.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when the agent processes the intel_packet in Phase 2.
Capability inventory: The agent has the capability to execute shell scripts (bash) and perform network-based delivery to external services.
Sanitization: No sanitization or verification steps are defined for the external data before it is processed by the reasoning layer or the Python engine.

geo-sentinel