The Agent Skills Directory

[NO_CODE]: The skill relies on an external shell script (~/.claude/skills/continuous-learning/evaluate-session.sh) to perform its core logic, but the script is not included in the repository. This prevents a complete verification of the skill's safety.
[COMMAND_EXECUTION]: The instructions direct the user to modify ~/.claude/settings.json to register a 'Stop' hook. This mechanism automatically executes shell commands at the conclusion of every agent session, creating a persistent execution path for unverified scripts.
[DATA_EXFILTRATION]: The skill's primary function is the automated extraction of information from 'full session transcripts.' This process involves reading potentially sensitive data such as API keys, proprietary code, or internal logic from the conversation history and storing it in a permanent local directory.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface (Category 8) by ingesting untrusted session data. Ingestion points: Full session transcripts containing user and agent messages. Boundary markers: None specified in the documentation or configuration. Capability inventory: File-system write access and shell command execution through the stop-hook mechanism. Sanitization: No validation or escaping of extracted session content is described before it is persisted as a 'learned' skill.

continuous-learning