The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Multiple files contain hardcoded sensitive credentials for the Feishu platform. Specifically, feishu_api.py includes REDACTED_FEISHU_PERSONAL_SECRET and REDACTED_FEISHU_OPENCLAW_SECRET, and scripts/md2feishu.sh contains a hardcoded APP_ID. While labeled as redacted, these placeholders in production code represent a significant risk of credential exposure and misuse if replaced with real values.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a diagnostic command !cat ~/.claude.json | grep -A 15 "lark-mcp" designed to execute at load time. This pattern accesses the agent's global configuration file, which may contain sensitive keys or setup information for other tools, posing a risk of silent metadata exfiltration.
[COMMAND_EXECUTION]: The skill provides scripts (feishu-mcp-setup.js and feishu-mcp-setup.sh) that programmatically modify the user's ~/.claude.json configuration file. This allows the skill to persistently alter how the agent executes tools and could be used to inject malicious tool configurations or intercept agent communications.
[DATA_EXFILTRATION]: Instructions in SKILL.md and logic in scripts/md2feishu.sh automate the process of uploading local Markdown files to a remote cloud platform (Feishu). If combined with malicious prompt instructions, this mechanism could be used to silently exfiltrate sensitive local documents from the user's environment.
[PRIVILEGE_ESCALATION]: The configuration setup scripts (feishu-mcp-setup.js) perform direct file system writes to the agent's core configuration directory (~/.claude.json). This behavior grants the skill significant control over the agent's execution environment beyond its stated purpose of Feishu automation.

feishu-automation