The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on npx @openfinclaw/cli@latest install, which downloads and executes the package directly from the npm registry. The source code repository linked (github.com/wikidjon/ai-openclaw-cli) is a personal repository that does not align with the official vendor namespace for 'reason-machines'.
[REMOTE_CODE_EXECUTION]: The installation process and the MCP server configuration (npx @openfinclaw/cli serve) involve the execution of remote scripts and binaries. This provides a mechanism for remote code execution under the guise of platform setup.
[DATA_EXFILTRATION]: The openfinclaw publish command enables the packaging of local directories into ZIP files and uploading them to hub.openfinclaw.ai. This is a documented feature for sharing trading strategies, but it creates a vector for data exfiltration if the agent is instructed to publish directories containing sensitive information like .env files or credentials.
[INDIRECT_PROMPT_INJECTION]: The platform processes community-contributed strategies downloaded via the strategy_fork command. These strategies consist of local files that the AI agent is instructed to read, modify, and validate, creating an attack surface for indirect prompt injection if the downloaded content contains malicious instructions for the LLM.
Ingestion points: Strategies downloaded from the leaderboard via openfinclaw fork; market data processed by deepagent tools.
Boundary markers: No specific boundary markers or instruction-guarding delimiters are identified for processing external strategy files.
Capability inventory: File system write/read (fork, validate, publish), network access (api, research_submit), and shell command execution through the CLI.
Sanitization: There is no evidence of sandboxing or static analysis performed on forked strategies before they are incorporated into the local workspace.
[DYNAMIC_EXECUTION]: The backtesting workflow involves the execution of local Python files (strategy.py). The skill encourages users to run these strategies through the platform's tools, which involves dynamic execution of the code defined within those scripts.
[COMMAND_EXECUTION]: The skill uses multiple commands that interact with the system shell, including installation scripts, environment diagnostics (openfinclaw doctor), and automated MCP server registration in configuration files (settings.json).

openfinclaw-quant-research