The Agent Skills Directory

[PROMPT_INJECTION]: Multiple reference documents, including references/architecture-patterns.md, references/blockchain-infrastructure-ref-1.md, and several others, contain extreme amounts of repetitive filler content. For example, references/blockchain-infrastructure-ref-1.md repeats the same technical topology description hundreds of times. This "context window bloating" technique can be used to hide malicious instructions in the noise or displace important security constraints from the agent's focus.
[PROMPT_INJECTION]: The skill documents components that ingest untrusted data from blockchain sources without clear boundary markers or sanitization, creating an indirect prompt injection surface.
Ingestion points: blockchain_exporter.py (reads w3.txpool.content) and the Forta bot example (reads TransactionEvent) in references/monitoring-alerting.md.
Boundary markers: Absent in the provided scripts.
Capability inventory: Network requests (web3) and finding generation for alerting systems.
Sanitization: Absent.
[COMMAND_EXECUTION]: The contract verification example in references/ci-cd-smart-contracts.md utilizes Foundry's Foreign Function Interface (vm.ffi) to execute shell commands directly from the testing runtime. While demonstrated for a legitimate purpose (running the forge CLI), this feature grants the execution environment high privileges and could be abused if inputs are not strictly controlled.

blockchain-infrastructure