The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/collect-data.sh script reads sensitive local data including full conversation history from ~/.claude/projects/ (in .jsonl format), memory summaries in MEMORY.md, and local configuration in .claude/settings.local.json. This exposure to the agent context is necessary for the skill's diagnostic purpose but involves accessing highly personal and configuration-sensitive information.
[COMMAND_EXECUTION]: The skill executes a shell script scripts/collect-data.sh to perform its audit. This script runs various system commands including git, find, jq, and python3 to aggregate environment data. It also uses npx skills path to resolve resource locations for the 'tw93/Waza' repository.
[PROMPT_INJECTION]: The skill processes untrusted data from previous conversations and external skill files. This creates an indirect prompt injection surface. The skill addresses this risk in agents/inspector-context.md by explicitly instructing the analyzing agent to treat the input bundle as untrusted and to ignore any instructions embedded within the audited content.

health