Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed business-idea research helper that uses background sessions, web search, local output files, and chat delivery in ways that fit its stated purpose.
Before installing, understand that saying `Idea: ...` starts a background research session, stores the submitted idea and generated report under `~/clawd/ideas/`, keeps the spawned session for later review, and sends results back through the chat. Avoid using highly confidential business ideas unless that storage and chat delivery are acceptable.
The README tells users to invoke the skill by saying "Idea: ..." but does not define whether that is the only accepted trigger, what variations are allowed, or what should not trigger it. It also defers to AGENTS.md for the actual trigger pattern, so this file's invocation guidance is incomplete and may lead to unintended or inconsistent activation.
The README states that research output is saved to '~/clawd/ideas//research.md' but does not warn users that their prompt contents and derived analysis will be persisted locally. Because business ideas may contain sensitive or proprietary information, silent local storage can create confidentiality and privacy risk, especially on shared systems or where home directories are backed up, synced, or monitored.
The manifest states the skill is rewritten to avoid external dependencies and to send results to the current chat, implying a Clawdbot-native workflow. However, the file repeatedly instructs sending files and summaries via Telegram chat, which reintroduces a platform-specific dependency and does not fully match the 'zero external dependencies' claim.
The skill causes user-supplied idea content and generated analysis to be written to local storage and sent back through chat, but it does not clearly warn users about this persistence and transmission. For potentially sensitive business ideas, this creates confidentiality and privacy risk because users may disclose proprietary information without understanding it will be stored on disk and redistributed through the messaging channel.
The prompt explicitly instructs the agent to create directories and write a research file under ~/clawd/ideas/{IDEA_SLUG}/research.md without indicating that this side effect must be disclosed to or confirmed by the user. In an agent setting, hidden filesystem writes can violate user expectations, create persistence, and become dangerous if the interpolated slug or surrounding workflow is later abused to overwrite unintended files.
The manifest emphasizes that results are sent to the current chat, distinguishing this version from the earlier Telegram/Saved Messages flow. The quick-start text explicitly says results are sent back to 'this Telegram chat,' which conflicts with the stated intent that this is a native current-chat implementation rather than a Telegram-centric one.
No suspicious patterns detected.
