The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches Twitter post data and historical snapshots from external sources including the Internet Archive (web.archive.org), Common Crawl (commoncrawl.org), and third-party renderers (api.fxtwitter.com and api.vxtwitter.com). These are well-known services or standard utilities for data recovery.
[PERSISTENCE]: Implements a cross-session user profile system using scripts/profile_store.py. This script stores user-stated preferences, brand information, and workspace settings in a local JSON file located at ~/.config/agent-skills/profile.json (or platform equivalents). It includes a denylist to prevent the storage of sensitive data such as tokens or passwords in the profile records.
[COMMAND_EXECUTION]: The workflow involves executing internal Python scripts (search_twitter.py and profile_store.py) to process status IDs, extract text from HTML/JSON responses, and register screenshot evidence. These actions are aligned with the skill's stated purpose of recovering verbatim X content.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests raw HTML and JSON data from external web sources and archives via scripts/search_twitter.py.
Boundary markers: The SKILL.md instructions explicitly require the agent to label verbatim text separately from translations or summaries, providing a clear boundary for the model to prevent accidental instruction following from external content.
Capability inventory: Capabilities include network access via the urllib library and local file writes to the profile configuration file.
Sanitization: The skill uses HTML parsing and regex extraction to isolate specific post content from the surrounding page source, rather than interpolating raw page data directly into prompts.

lov-search-twitter