The Agent Skills Directory

[SAFE]: The skill contains legitimate technical documentation and code examples for managing LangGraph checkpoints. No malicious code, obfuscation, or unauthorized data access patterns were identified.
[INDIRECT_PROMPT_INJECTION]: The instructions describe building stateful agents that process user-provided messages. The skill includes security-focused advice to mitigate risks, such as enforcing tenant-specific thread IDs to prevent data leaks between users and implementing human-in-the-loop interrupts to validate sensitive state transitions.
Ingestion points: User messages ingested via graph.invoke as shown in SKILL.md and references/thread-id-discipline.md.
Boundary markers: The skill recommends the use of interrupt_before (Step 4 in SKILL.md) to implement human-in-the-loop validation.
Capability inventory: Examples in SKILL.md show the agent using tools via create_react_agent.
Sanitization: The skill focuses on ensuring state is JSON-serializable and provides strategies for tenant isolation at the thread_id level.
[EXTERNAL_DOWNLOADS]: The skill references standard Python libraries (langgraph, langchain-core, langgraph-checkpoint-postgres, asyncpg) for its core functionality. These are well-known packages in the LangChain ecosystem.
[CREDENTIALS_UNSAFE]: The skill correctly demonstrates using environment variables (os.environ["DATABASE_URL"]) for database connection strings rather than hardcoding sensitive credentials.

langchain-langgraph-checkpointing