The Agent Skills Directory

[COMMAND_EXECUTION]: The utility scripts/find-living-plan.sh passes user-controlled slug arguments directly to the grep command, which creates a potential surface for argument injection if inputs are maliciously crafted.
[DYNAMIC_EXECUTION]: The file scripts/find-living-plan.sh employs a Bash heredoc to execute an embedded Python script (`python3
"$1" <<'PY'`) for parsing JSON metadata from generated HTML artifacts.
[EXTERNAL_DOWNLOADS]: The skill integrates with Google's NotebookLM via a specialized MCP tool (mcp__notebooklm-mcp__studio_create) to upload plan briefs and generate infographics or slides. As a well-known service from a trusted provider, this interaction is considered a standard capability for its stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external codebases and git diffs to generate architectural visualizations, creating a potential vector for indirect prompt injection if the analyzed repository contains malicious instructions.
Ingestion points: Code and diff data are ingested in scripts/analyze-impact.sh and the Explore sub-agent prompt in SKILL.md.
Boundary markers: Absent; there are no specific delimiters used to isolate processed repository content from the agent's instructions.
Capability inventory: The skill uses Bash, Write, and sub-agents (Agent) as defined in the frontmatter of SKILL.md.
Sanitization: Absent; the skill does not filter or sanitize instructions that may be embedded in commit messages or source code comments.

visualize-plan