The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill frequently uses npx @clawlogic/sdk@latest within its documentation and setup instructions. This practice fetches the most recent version of the SDK from the NPM registry at runtime, representing an unversioned external dependency that could change behavior without notice.
[COMMAND_EXECUTION]: The skill's architecture relies on shell scripts (scripts/*.sh) that execute TypeScript helpers via npx tsx. This pattern involves executing code from the local filesystem and managing environment variables like AGENT_PRIVATE_KEY for transaction signing.
[DATA_EXFILTRATION]: The scripts/helpers/post-broadcast.ts utility performs network POST operations to a destination defined by the AGENT_BROADCAST_URL environment variable (defaulting to a local endpoint). This constitutes a network egress point that could be redirected to an external server to transmit agent rationale or metadata.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) by ingesting untrusted data from the blockchain.
Ingestion points: The scripts/helpers/analyze-market.ts script fetches market description, outcome1, and outcome2 labels from the CLAWLOGIC smart contracts.
Boundary markers: No boundary markers or 'ignore' instructions are used when passing these strings into the agent's reasoning context.
Capability inventory: The agent has the capability to execute financial transactions on-chain (buy, mint, assert) and perform external network requests (broadcast).
Sanitization: There is no evidence of sanitization or validation of the market description string, allowing an attacker who creates a market to potentially influence the agent's trading logic.

clawlogic-trader