The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted marketplace data from the LienFi API while maintaining the capability to execute blockchain transactions.\n
Ingestion points: The agent retrieves JSON data from api.lienfi.com via tools such as search_liens and get_lien as described in SKILL.md.\n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instruction' warnings when processing the retrieved marketplace data.\n
Capability inventory: The agent can request signatures (/wallet/sign) and submit transactions (/wallet/submit) through the Bankr API, as outlined in the purchase lifecycle steps.\n
Sanitization: No explicit sanitization or validation of the remote marketplace content is mentioned before it is used to parameterize transactions.\n
Mitigation: The skill enforces a mandatory human authorization step where the operator must sign a bearer token in a browser, which includes enforcing spend caps.\n- [COMMAND_EXECUTION]: The skill relies on shell commands for all network and service interactions.\n
Details: The agent is instructed to use curl to communicate with api.lienfi.com and api.bankr.bot. These operations are functional requirements for the skill's primary purpose and target either vendor-owned or service-specific domains.\n- [PRIVILEGE_ESCALATION]: The setup instructions require the operator to configure the wallet API key with broad permissions.\n
Details: SKILL.md and catalog.json state that the BANKR_API_KEY must have allowedRecipients set to EMPTY and readOnly set to OFF. This configuration allows the agent to sign and submit transactions to any contract, which is a documented prerequisite for the agent's DeFi operations but represents a reduction in default security constraints.

lienfi