The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and summarize content from social media platforms, creating an attack surface for indirect prompt injection.
Ingestion points: Scraper scripts (e.g., scripts/engine/sources/aside_x.py, scripts/engine/sources/reddit_keyless.py, scripts/engine/sources/youtube_ytdlp.py) fetch untrusted text from X, Reddit, YouTube, and other feeds.
Boundary markers: SKILL.md contains an explicit warning instructing the agent to treat all scraped text as untrusted content and to never follow instructions found within it.
Capability inventory: The skill possesses the capability to execute shell commands via subprocess.run (found in scripts/engine/backend.py and scripts/engine/sources/youtube_ytdlp.py) and perform network requests via urllib.request (found in scripts/engine/httpget.py).
Sanitization: The skill processes fetched data and provides it to the agent; while the agent is warned, the raw text from external sources is not explicitly sanitized of potential prompt-overriding content before presentation.
[COMMAND_EXECUTION]: The skill wraps and executes several external binaries to perform data extraction tasks.
Evidence: scripts/engine/backend.py executes the aside CLI to run JavaScript for browser-based scraping. scripts/engine/sources/youtube_ytdlp.py executes the yt-dlp binary, and scripts/engine/sources/github_issues.py executes the GitHub CLI (gh).
Implementation: These executions use the list-based subprocess.run API without a shell (shell=False), which follows security best practices for executing system commands by preventing shell metacharacter injection.

last30days-local