Agent Memory and Recovery (SQLite MVP)

This documents a simple local SQLite strategy to help agents avoid repeated failures and recover from broken design states.

Goal

Capture enough execution history to answer:

1. What failed?
2. Where did it fail (tool + stage)?
3. What is the likely root cause?
4. What should we do next?
5. Should we roll back before retrying?

Local Storage

Database path:

• .solidworks_mcp/agent_memory.sqlite3

Core tables:

• agent_runs
• tool_events
• error_catalog

Error-Centric Recovery Loop

1. Query error_catalog for similar failures by tool_name + error_type.
2. Present the top remediation options to the user.
3. Prefer rollback/recovery actions before blind retries.
4. Re-run with constrained parameters and capture outcome.

Example Troubleshooting Prompt

"Based on recent create_extrusion failures in the local SQLite catalog, suggest rollback-first recovery using existing MCP tools and avoid repeating known bad states."

Rollback-First Tooling Pattern

When a step fails:

1. Capture snapshot metadata.
2. Inspect recent feature-tree changes.
3. Roll back/suppress latest risky feature.
4. Re-run prerequisite validation tools.
5. Retry only after preconditions pass.

Why This Helps

• Fewer repeated failures
• Faster root-cause diagnosis
• Better agent recommendations over time
• Better handoff between human and agent sessions