The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it extracts persistent rules from untrusted text samples.
Ingestion points: Processes writing samples provided by the user to capture voice summaries and style constraints (references/writing-style-learning.md).
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the analyzed samples.
Capability inventory: Extracted style rules directly influence the agent's drafting behavior and personalization output (SKILL.md).
Sanitization: The skill lacks technical sanitization for ingested text, though it notes best practices to avoid learning from boilerplate or signatures.
[REMOTE_CODE_EXECUTION]: Documentation in SKILL.md instructs users to execute a Python script (scripts/install-external-skills.py) to install external components. The script's logic is not available in the skill package for verification.
[EXTERNAL_DOWNLOADS]: The skill references an external registry located at ../../../references/external-skills.yaml and lists external skills like unslop and stop-slop for installation via the aforementioned script.

communication-style