The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the extraction and persistence of information from conversation history into local snapshot files (data/*.md) and the project's MEMORY.md file. This creates an indirect prompt injection surface where malicious instructions embedded in user input could be stored and subsequently re-injected into the agent's context during later sessions.
Ingestion points: The entire conversation history is parsed for extraction during Phase 1 (Extracao Estruturada).
Boundary markers: The skill uses markdown headers, tables, and specific markers (e.g., ) to structure extracted content.
Capability inventory: The skill possesses file-write capabilities (via context_snapshot.py) and execution capabilities (via script calls to manage snapshots).
Sanitization: No explicit sanitization, validation, or escaping of the extracted external content is implemented; the skill relies on the agent's categorization logic.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Python scripts to perform data saving and management tasks. This includes context_snapshot.py within the skill directory and a script context_manager.py from a separate, assumed-to-be-present skill named context-agent at a specific absolute path (C:\Users\renat\skills\context-agent\).

context-guardian