The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local project data, including files, documentation, and git history, to understand user intent. This creates a surface for indirect prompt injection if those files contain malicious instructions.\n
Ingestion points: Instructions in SKILL.md direct the agent to "Explore project context — check files, docs, recent commits".\n
Boundary markers: No specific delimiters or warnings to ignore embedded instructions are present in the skill's data processing flow.\n
Capability inventory: The skill can write files and invoke the writing-plans skill for further implementation steps.\n
Sanitization: There is no evidence of sanitization or filtering applied to the content read from the project context.\n- [COMMAND_EXECUTION]: The skill spawns subprocesses to open the browser-based companion interface.\n
Evidence: scripts/server.cjs uses execFile to launch platform-specific commands like open, xdg-open, or rundll32.exe.\n
Evidence: It also supports a customizable launcher via the BRAINSTORM_OPEN_CMD environment variable, which is executed using the less-secure exec function.\n- [EXTERNAL_DOWNLOADS]: The visual companion interface loads a branding image from an external vendor domain.\n
Evidence: scripts/server.cjs references https://primeradiant.com/brand/superpowers-visual-brainstorming-logo.png for use as a brand logo.

brainstorming