The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a detailed schema for ingesting and parsing Claude Code session transcripts, which represents an attack surface where historical data can influence current agent behavior.
Ingestion points: The documentation specifies the location and structure of session logs at ~/.claude/projects/**/*.jsonl and related task files in ~/.claude/tasks/ as primary sources of data.
Boundary markers: The schema defines structural boundaries via record types (e.g., user, assistant, system), but it does not provide mechanisms to distinguish between data content and embedded instructions within those fields.
Capability inventory: The skill is a static documentation reference and does not include any executable scripts, tool calls, or network operations.
Sanitization: The reference does not provide logic for sanitizing, escaping, or filtering the raw content strings found in the logs before they are interpreted by an agent.
[SAFE]: External references to the langmartai/lm-assist GitHub repository are provided solely for documentation and schema verification purposes. No automated downloads or execution of this third-party code are performed by the skill.

claude-session-data-schema-reference