The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process untrusted data from other skill executions, including raw user queries and skill outputs.
Ingestion points: Data enters the system via the user_query and output fields processed in SKILL.md (e.g., in log_skill_invocation and identify_improvement_opportunities).
Boundary markers: No specific delimiters or instructions to treat ingested strings as data rather than instructions are present in the logic.
Capability inventory: The skill is granted powerful capabilities including Bash, Write, Edit, and Read access to the filesystem.
Sanitization: The provided implementation lacks sanitization or escaping of the ingested strings before they are stored or potentially re-processed by the agent.
[DATA_EXFILTRATION]: The skill implements a logging mechanism that stores detailed interaction history, including full user prompts and assistant responses, in a local SQLite database located at ~/.claude/skill_logs.db. This creates a centralized repository of potentially sensitive information on the local filesystem.

skill-logger