The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data retrieved from Twitter/X via the CLI (e.g., drafts, plans, and outcomes). This represents a potential surface for indirect prompt injection if external content contains malicious instructions.
Ingestion Points: Output from tw plan review list, tw task review list, and tw draft show commands.
Boundary Markers: The skill explicitly instructs the agent to pause for human review before every approval, rejection, or execution step, serving as a primary defense.
Capability Inventory: Capabilities include content generation, tweet scheduling, and strategy activation via the tw CLI.
Sanitization: Metadata is sanitized before being included in issue reports, and private references are excluded from recovery summaries to prevent data leakage.
[COMMAND_EXECUTION]: The skill makes extensive use of a local CLI tool (tw) to manage Twitter operations. These commands are executed locally and are restricted to specific capability families defined in the manifest (context, strategy, plan, task, draft, scheduler).
[EXTERNAL_DOWNLOADS]: The skill references external URLs for setup and updates.
Evidence: Fetches setup instructions from https://www.threadwave.xyz/cli/setup/agent.md and checks for updates via https://raw.githubusercontent.com/ohmyskyhigh/threadwave-skill/main/release-index.json.
Context: These resources are hosted on the vendor's official domain and GitHub repository, representing standard operational functionality.

twitter-agent