The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The script evaluate-session.sh executes standard local utilities including mkdir, grep, and jq. These are used for directory management and session metadata analysis. No evidence of arbitrary command execution or shell injection from external sources was found.
[DATA_EXPOSURE] (SAFE): The skill interacts with the user's home directory (~/.claude/skills/learned/) to store extracted patterns. This is limited to the tool's intended workspace and does not involve access to sensitive system files, credentials, or network transmission.
[INDIRECT_PROMPT_INJECTION] (SAFE): The skill ingests session transcripts to evaluate learning opportunities.
Ingestion points: The script reads session transcripts via the path provided in the CLAUDE_TRANSCRIPT_PATH environment variable.
Boundary markers: None used; the script processes the raw transcript file.
Capability inventory: Directory creation (mkdir), JSON parsing (jq), and pattern counting (grep). The script lacks capabilities for network communication or code execution.
Sanitization: The transcript data is only used as input for grep -c to count messages, ensuring that any malicious instructions within the transcript are not executed or interpreted by the shell.

agent-continuous-learning