The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including interview transcripts, source packs, and notes, which creates a surface for indirect prompt injection.
Ingestion points: SKILL.md specifies that the agent accepts topics, briefs, notes, interviews, transcripts, and source packs as valid inputs.
Boundary markers: The skill includes a robust references/source-policy.md which instructs the agent to separate verified facts from interpretations and user preferences, and to build a source map before drafting.
Capability inventory: The skill's primary capability is text generation (articles, guides, outlines). It does not explicitly request high-privilege tools like subprocess execution or arbitrary file writing.
Sanitization: references/final-qa.md mandates a source integrity check to ensure all facts, numbers, and quotations are traceable to the source map.
[EXTERNAL_DOWNLOADS]: The skill references an external Telegram channel (https://t.me/a1_marketing_skills) for user support and feedback. Telegram is a well-known service, and the link is used for legitimate support purposes.
[DATA_EXFILTRATION]: The skill includes instructions to read project-specific marketing context from files such as .agents/marketing-context.md, .claude/marketing-context.md, and marketing-context.md. This is a standard practice for repository-aware agents to adapt to specific project guidelines and does not target sensitive system credentials or user data.

a1-article-writing