The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git commands (git status, git diff, git commit) to maintain the local content repository. These operations are intended for managing the lifecycle of generated drafts and do not involve system-level privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically X/Twitter posts and bookmarks. This constitutes an injection surface where untrusted content could potentially influence the agent's behavior. The skill instructions address this by specifying structural extraction and voice calibration patterns, which focus the agent on data processing rather than instruction following from the ingested text.
Ingestion points: Browser navigation to x.com search results and bookmark feeds; local reading of AGENTS.md and existing drafts.
Boundary markers: The agent is instructed to distinguish between authored text and quoted sources, and to focus on measurable patterns (hook usage, paragraph length) rather than copying fragments.
Capability inventory: File system access for reading/writing drafts; shell command execution via git; browser access for viewing X.com.
Sanitization: No specific content sanitization or escaping is mentioned for the ingested social media text.
[EXTERNAL_DOWNLOADS]: The skill uses the in-app browser to fetch data from x.com, a well-known service. These operations are scoped to public search and authenticated bookmark views necessary for the skill's stated purpose.

x-bookmark-quote-posts