The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources, creating a surface for indirect prompt injection where malicious instructions embedded in project files or issues could influence agent behavior.
Ingestion points: The skill ingests data from user arguments ($ARGUMENTS), git diff output via scripts/analyze-impact.sh, GitHub issue content via gh issue view, and project file contents accessed via the Read tool.
Boundary markers: Prompt templates, such as those used for the Explore sub-agent, do not explicitly employ delimiters or clear instructions to treat the interpolated project data as non-executable text.
Capability inventory: The skill possesses significant capabilities including Bash for command execution, Write for file modification, Agent for spawning sub-agents, and mcp__notebooklm-mcp__studio_create for external tool interaction.
Sanitization: There is no evidence of specific sanitization, escaping, or validation of the external data before it is interpolated into prompts or task subjects (e.g., TaskCreate(subject="Visualize plan: {PLAN_INPUT}", ...)).
[COMMAND_EXECUTION]: The skill relies on several local scripts to analyze the environment and project state.
Evidence: It executes scripts/analyze-impact.sh, scripts/detect-plan-context.sh, and scripts/find-living-plan.sh using the Bash tool. These scripts further invoke standard utilities like git and the gh (GitHub CLI).
[SAFE]: The skill's interactions with external services are limited to well-known and trusted platforms.
Evidence: The use of gh issue view for GitHub integration and mcp__notebooklm-mcp__studio_create for NotebookLM functionality involves established services. The skill's metadata and directory structure are consistent with its stated purpose of architectural visualization and plan management.

visualize-plan