The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses highly sensitive local data through the following mechanisms:
- It executes grep against ~/.claude/memory/ to retrieve historical user context and prior decisions.
- It accesses and reads from ${MAKERSKILLS_CONFIG:-$HOME/.config/makerskills}/deep-research/archive/.
- The skill simultaneously uses network-enabled tools (WebSearch, WebFetch, agent-browser) which could potentially receive or process fragments of this sensitive local data during the synthesis phase.
[COMMAND_EXECUTION]: The skill utilizes shell-level commands and filesystem operations to manage its research corpus:
- Uses grep for searching local directories (~/.claude/memory/ and archive paths).
- Performs filesystem writes to save research briefs to ~/.config/makerskills/deep-research/archive/.
- Instructs a file migration operation: "move those files into the archive directory first" if an old structure exists.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its core function of processing untrusted web data:
- Ingestion points: Untrusted data enters the agent via WebSearch, WebFetch, and the agent-browser tool which hits arbitrary URLs.
- Boundary markers: There are no instructions provided to wrap external content in delimiters or to ignore potential instructions embedded in the fetched web pages.
- Capability inventory: The agent has the ability to read private user memory via grep, interact with the Notion API (using $NOTION_API_KEY), and write to the local filesystem.
- Sanitization: The instructions do not define any sanitization or validation steps for content retrieved from the web before it is synthesized into the final brief.

deep-research