The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses internal platform history and session metadata located at ~/.claude/commander/knowledge/. This directory contains sensitive logs of previous user interactions, tool outputs, and environment states used for learning.
[COMMAND_EXECUTION]: Executes shell commands via the Bash tool to perform file system inspection and count historical records (ls ~/.claude/commander/knowledge/*.json | wc -l).
[PROMPT_INJECTION]: Susceptible to Indirect Prompt Injection (Category 8). The skill processes data from previous sessions which may include untrusted content (e.g., outputs from external tools or user-provided data). If an attacker successfully injects a malicious pattern into the session history, this skill may interpret it as a "successful approach" and propose its inclusion in other SKILL.md files.
Ingestion points: Reads all JSON files within ~/.claude/commander/knowledge/ (SKILL.md).
Boundary markers: None; the skill extracts logic directly from historical JSON payloads without verification of origin.
Capability inventory: Possesses Read, Write, and Edit permissions for all skill files, and Bash for system commands (SKILL.md).
Sanitization: No automated sanitization or filtering of historical patterns; relies entirely on the user to identify malicious content during the AskUserQuestion approval step.

skill-evolution