The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the ~/.claude/memory/ directory to retrieve prior research and context. This path is used by the agent to store sensitive history of user interactions and persona data.
[CREDENTIALS_UNSAFE]: The skill instructs the agent to utilize a sensitive environment variable, $NOTION_API_KEY, for authentication and data retrieval from external Notion workspaces.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and synthesizes data from untrusted external sources.
Ingestion points: Data is fetched from arbitrary URLs, Google search results, and social media platforms (Reddit, X, YouTube, HN) via WebSearch, WebFetch, and the agent-browser tool.
Boundary markers: The instructions do not provide explicit delimiters or "ignore embedded instructions" directives to prevent the agent from following malicious commands hidden in the fetched web content.
Capability inventory: The agent has capabilities to perform file system operations (reading/writing to ~/.config/makerskills/), execute shell searches (grep), and communicate with the external Notion API.
Sanitization: No validation or sanitization is performed on the content retrieved from external sources before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: Performs network requests to non-whitelisted domains to fetch web content and search results.
[COMMAND_EXECUTION]: The skill uses shell-level commands like grep to search local memory and archive files, and creates directories using mkdir at ~/.config/makerskills/deep-research/archive/.

deep-research