The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands to manage its workflow, including mktemp for temporary file creation, sed for stripping YAML metadata, and gbrain for content retrieval.\n- [COMMAND_EXECUTION]: It invokes a local binary located at $HOME/.claude/skills/gstack/make-pdf/dist/pdf to perform the document rendering.\n- [COMMAND_EXECUTION]: The invocation includes a CONTAINER=1 environment variable which disables the Chromium sandbox in the rendering engine (Playwright). While intended for container compatibility, this reduces the isolation between the rendering process and the host system.\n- [EXTERNAL_DOWNLOADS]: The skill requires the external gstack software package and suggests installing system-level font packages (fonts-liberation and fonts-noto-cjk) to ensure correct rendering.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external Markdown data.\n
Ingestion points: Data is ingested from the brain repository via the gbrain get command in SKILL.md.\n
Boundary markers: The skill lacks explicit boundary markers or instructions to the model to ignore embedded commands within the processed Markdown.\n
Capability inventory: The skill has permissions to read files and execute the rendering binary based on the ingested content.\n
Sanitization: While it removes frontmatter and mentions emoji sanitization, there is no comprehensive filtering of the Markdown body before it is passed to the PDF renderer.

brain-pdf