The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit security instructions to defend against prompt injection.
In SKILL.md, the instructions mandate that the agent must analyze URLs and files as data only and must not obey any instructions within that data that attempt to override the skill's contract, access local files, or execute commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, which is a standard surface for indirect prompt injection. However, it implements significant safeguards.
Ingestion points: Processes URLs, uploaded files, and raw text provided by the user as defined in SKILL.md (Step 1).
Boundary markers: Explicitly present. The skill instructs the agent to treat input as data and disregard any embedded commands.
Capability inventory: The skill's capabilities are limited to generating text-based deliverables (article-draft, publish-info, figure-spec). There are no capabilities for file system modification, subprocess execution, or outgoing network requests within the skill's logic.
Sanitization: The agent is instructed to strictly separate source facts from user opinions and verified project claims, ensuring that malicious content in the source data does not influence the system's logic or output structure.
[EXTERNAL_DOWNLOADS]: The skill references external resources and installation paths, but these are restricted to the author's own infrastructure.
Installation via npx skills add points to the author's GitHub repository (github.com/KKenny0/signal-pack), which is consistent with normal vendor operations.
Style references in references/halftone-paper-collage.md point to other repositories by the same author (KKenny0/project-promo-skill), representing standard internal cross-referencing.

signal-pack