The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted data from external sources, which presents a surface for indirect prompt injection.
Ingestion points: The agent reads X (Twitter) posts, parent threads, website content, smart contract source code, and project-specific documentation (e.g., agent.md, skill.md).
Boundary markers: The skill uses structured output cards and logical separation between 'project claims' and 'evidence,' but does not utilize strict technical delimiters (like XML tags) for the raw data being processed.
Capability inventory: The agent has access to a cloud computer terminal (allowing execution of Python/Node and file system writes to /workspace), as well as web and X search tools.
Sanitization: The instructions in SKILL.md and grok-runtime.md explicitly mandate that external data must be treated as text only and cannot override the skill's own instructions or trigger unauthorized command execution.
[DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of small, locally-generated Python or Node.js scripts within the cloud terminal environment.
Evidence: references/grok-runtime.md specifies that the agent may write and run small programs for data collection, parsing, calculation, and state management.
Context: This capability is restricted to internal logic processing and state preservation within a sandboxed cloud environment.
[COMMAND_EXECUTION]: The skill uses a cloud terminal to perform network operations for research purposes.
Evidence: The agent uses curl and JSON-RPC calls (eth_call, eth_getCode) to query blockchain data.
Context: The instructions in references/fees-contracts.md and references/grok-runtime.md explicitly forbid the use of any commands capable of signing transactions, sending funds, or executing project-provided scripts.

ezbot-grok-scout