The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, user-provided content such as raw notes, transcripts, and results to generate threads. This creates a surface for indirect prompt injection if the source material contains malicious instructions designed to influence the agent's behavior during the drafting process.
Ingestion points: Topic and raw material (notes, transcript, result) gathered in Step 1 of SKILL.md.
Boundary markers: The skill does not explicitly define delimiters for user input, though it enforces strict formatting and human review steps.
Capability inventory: The skill has the capability to publish content via lib.publish in SKILL.md.
Sanitization: No explicit sanitization or filtering of external content is mentioned before the thread generation phase, however, a mandatory human-in-the-loop approval step (Step 9) is required before any publishing occurs.
[SAFE]: The skill uses an internal library tool (lib.publish) to interact with a well-known service (https://x.com/compose/tweet) for its primary intended purpose. This interaction is transparently documented in the skill instructions.
[SAFE]: The skill references local configuration files (e.g., voice-profile.md, story-bank.md) to maintain consistency. These references are static and do not involve user-controlled paths or external network requests.

x-thread-builder