The Agent Skills Directory

[PROMPT_INJECTION]: The skill's workflow (Steps 2 and 3 in SKILL.md) instructs the agent to analyze user-provided file content to generate summaries, highlights, and optimized titles. This ingestion of untrusted data lacks boundary markers or instructions to ignore embedded commands, creating a surface for indirect prompt injection. Ingestion points: Content read from user-specified markdown or plain text files. Boundary markers: None identified in the instructional prompts. Capability inventory: Includes file write operations and shell command execution. Sanitization: Input content is processed without prior validation or escaping of potential injection payloads.
[EXTERNAL_DOWNLOADS]: During the typography fix phase (Step 6), the skill uses npx to execute the autocorrect-node package. This triggers a download from the npm registry at runtime if the package is not cached locally.
[COMMAND_EXECUTION]: The skill leverages node:child_process.spawnSync to execute system commands (npx or bun) for processing the output files. While used for formatting, this provides a mechanism for interacting with the host shell environment. The main entry point also includes a function to decode hex-encoded HTML entities, which is a common technique for bypassing simple pattern-based filters.

baoyu-format-markdown