The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses session transcripts located in ~/.claude/projects/**/*.jsonl and ~/.codex/sessions/**/*.jsonl. These files contain historical conversation data, including user messages and assistant responses, which may contain sensitive personal information or credentials previously shared during sessions.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external skills and session logs without explicit sanitization or strict boundary markers. * Ingestion points: Reads SKILL.md files and JSONL session logs. * Boundary markers: None identified in the provided instructions. * Capability inventory: Uses bash, python3, test -e, and which. * Sanitization: None identified in the provided instructions.
[COMMAND_EXECUTION]: Uses shell commands including bash, test -e, and which to verify the existence of files and CLI tools in the local environment as part of its environment consistency check.
[DYNAMIC_EXECUTION]: Instructs the agent to execute python3 scripts via bash to scan and process local session transcripts, involving runtime execution of code to perform diagnostics.

skill-optimizer