wind-mcp-research-skill
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPERSISTENCEDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [METADATA_POISONING]: The skill's frontmatter in
SKILL.mdclaims thatchild_processusage isfalse, but the implementation inscripts/cli.mjsandscripts/update-check.mjsrelies heavily on spawning child processes for background updates and system checks. - [PERSISTENCE]: The skill implements a background update check in
scripts/cli.mjsthat triggers after successful tool calls. It spawns a detached process (scripts/update-check.mjs) that runs independently of the main agent session to maintain and update the skill's files daily. - [COMMAND_EXECUTION]: The skill executes shell commands via
node:child_process.scripts/cli.mjsspawns background update tasks, andscripts/update-check.mjsexecutesgit,npx skills update, andnpx skills addto modify the local environment. - [DYNAMIC_EXECUTION]: The update mechanism in
scripts/cli.mjscopies the update script to a temporary cache directory (e.g.,~/.cache/wind-aifinmarket/) and executes it dynamically to avoid conflicts during the self-update process. - [REMOTE_CODE_EXECUTION]: The skill is configured to automatically pull and install code updates from remote repositories (GitHub and Gitee) using the
npx skills updatecommand. While these target vendor-owned repositories, this mechanism allows for remote modification of the skill's executable logic. - [INDIRECT_PROMPT_INJECTION]: The skill processes external financial data, news, and research reports which could contain adversarial instructions.
- Ingestion points: Financial data and documents retrieved from
https://mcp.wind.com.cnvia various MCP tools. - Boundary markers: The
SKILL.mdfile contains instructions for the agent to only base its answers on the returned tool results and avoid using internal knowledge. - Capability inventory: The skill has the ability to read and write to the file system (for configuration and updates), make network requests, and spawn background processes.
- Sanitization:
scripts/cli.mjsperforms parameter validation (types, enums, and required fields) against a local registry schema before executing tool calls.
Audit Metadata