The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches blockchain and task-related data from external network endpoints.
Evidence: Connects to base-rpc.publicnode.com to interact with the Base mainnet via RPC calls.
Evidence: Communicates with azzle.org APIs to retrieve marketplace metadata and task listings.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the task marketplace, such as task descriptions and scopes.
Ingestion points: Untrusted task metadata and off-chain messages (XMTP) are processed in scripts/v2-lib.mjs and used to populate the agent's context.
Boundary markers: SKILL.md provides explicit instructions to the agent to treat all marketplace data as untrusted and never to use it to authorize code execution or transactions.
Capability inventory: The skill has the ability to execute on-chain transactions and bounded approvals via the bankr tool.
Sanitization: The skill implements a robust defense-in-depth strategy, requiring the agent to verify all contract addresses, runtime code hashes, and implementation code hashes against a reviewed 'pinned' configuration before every write.
[COMMAND_EXECUTION]: The skill uses local shell scripts and CLI tools to verify blockchain state and execute transactions.
Evidence: Executes ./scripts/v2-tasks.sh to perform integrity checks on deployed contracts.
Evidence: Uses the bankr CLI for wallet operations and transaction signing, with mandatory local decoding and user confirmation steps defined in the instructions.

azzle