The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an automated context persistence mechanism that is vulnerable to indirect prompt injection.
Ingestion points: session_parser.py reads JSONL logs from the Claude Code projects directory (C:\Users\renat\.claude\projects).
Boundary markers: Extracted content is written to ACTIVE_CONTEXT.md and MEMORY.md without delimiters or 'ignore embedded instructions' warnings.
Capability inventory: The skill has file-write access via active_context.py and project_registry.py, and provides a full-text search interface using SQLite FTS5 in search.py.
Sanitization: No escaping, validation, or filtering is performed on message content during extraction in session_summary.py before it is saved to the active context and synced to the agent's memory file.
[COMMAND_EXECUTION]: The skill requires the manual execution of local Python scripts to maintain session state.
SKILL.md provides instructions to run several commands via python C:\Users\renat\skills\context-agent\scripts\context_manager.py, including init, save, and load commands.
These scripts interact with the file system to manage the data/ directory and synchronize information with the agent's environment.

context-agent