The Agent Skills Directory

[PROMPT_INJECTION]: The skill proactively addresses indirect prompt injection vulnerabilities through explicit instructions in SKILL.md and references/feed-leaderboard.md.
Ingestion points: Untrusted data enters the agent context via the public trade feed (/feed), trader lab entries (/lab/:wallet), user comments (/comments/:thesisId), and external news RSS feeds (api.rss2json.com).
Boundary markers: The skill uses a dedicated 'FORBIDDEN' section to instruct the agent to ignore any natural language commands (e.g., 'ignore previous instructions') found in untrusted content.
Capability inventory: High-privilege actions include market trades (/trade), fund withdrawals (/proxy/bankr-withdraw), and autonomous agent activation (/bankr/activate).
Sanitization: The agent is strictly instructed to only render or summarize external content, never to use it as a trigger for signing, credential disclosure, or order execution.
[DATA_EXFILTRATION]: Authenticated actions require a walletSig and bankrApiKey to be sent to the platform's backend (og.nexustradinglabs.com). The skill includes robust safety guidelines to prevent credential leaks, specifically forbidding the persistence, logging, or transcription of these session-based credentials. This represents standard vendor functionality for a non-custodial trading proxy.
[COMMAND_EXECUTION]: The skill manages financial transactions including leveraged trades and USDC withdrawals. These are protected by mandatory session signatures (sign_message) and a non-negotiable requirement for explicit user confirmation ('GO LIVE') before deploying autonomous agents with real funds.
[EXTERNAL_DOWNLOADS]: The skill retrieves market intelligence, funding rates, and news headlines from established sources like api-evm.orderly.org and api.rss2json.com. These are legitimate data fetches relevant to the skill's purpose and do not involve executable code or script injection.

nexus