The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Twitter/X (such as public posts, user biographies, and replies) to generate content drafts, research briefs, and suggested replies. This creates a surface for indirect prompt injection if malicious instructions are embedded in the retrieved X data.
Ingestion points: Commands like superx x:post, superx x:user, and superx engage:posts retrieve external data into the agent's context (e.g., in references/skills/who-is-this-person/recipe.md).
Capability inventory: The skill possesses capabilities to write to the local file system (superx datasets:export), schedule content (superx scheduled:create), and queue direct messages (superx dm:campaign).
Sanitization: The instructions do not specify explicit sanitization or escaping of the retrieved X data before it is processed by the AI.
Boundary markers: No specific delimiters or boundary markers are instructed for isolating untrusted data.
Mitigation: The skill strongly mitigates this risk by mandating human review and confirmation for all write operations. Rule 3 in SKILL.md and the 'Stops at' sections in every recipe (e.g., references/skills/thread-builder/recipe.md) emphasize that nothing is posted or sent without explicit human approval.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the superx-cli package from the official npm registry and refers to vendor-owned resources on GitHub.
Evidence: npm install -g superx-cli and links to github.com/superx-so/superx-agent in SKILL.md. These resources originate from the verified vendor for this skill.
[COMMAND_EXECUTION]: The skill's primary function is to orchestrate the execution of the superx CLI tool for managing Twitter/X growth activities.
Evidence: Extensive documentation of command usage for managing accounts, posts, analytics, and lead generation throughout SKILL.md and references/commands.md.

superx