The Agent Skills Directory

[DYNAMIC_EXECUTION]: The utility scripts scripts/migrate_state.py and scripts/test_reducers.py utilize importlib.util to dynamically load and execute code from local files provided as command-line arguments. This allows the user to test their own reducer functions and migrate existing checkpoint databases to new state schemas. While this involves executing arbitrary local code, it is the primary intended function of these development tools.
[COMMAND_EXECUTION]: The documentation instructs users to execute management tasks and validation checks using the uv run command. This represents standard local shell command execution for Python development workflows.
[INDIRECT_PROMPT_INJECTION]: The skill implements state management patterns that ingest untrusted data. 1. Ingestion points: Untrusted data enters the agent context through the messages field in MessagesState defined in assets/chat_state.py and references/schema-patterns.md. 2. Boundary markers: The skill does not explicitly include prompt-level boundary markers or instructions to ignore embedded instructions within its schema templates. 3. Capability inventory: The skill includes scripts scripts/migrate_state.py and scripts/inspect_checkpoints.py which perform file system read/write operations on SQLite databases, and the graph nodes can execute logic based on the state. 4. Sanitization: The skill uses the add_messages reducer which manages message history but does not perform content-level sanitization or filtering of the message content itself.

langgraph-state-management