- Home
- Skills
- Data & Databases
- Aegis-SQL-Bridge
More screenshots
Works with the AI Agents you already use
Aegis-SQL-Bridge
A self-healing, zero-trust SQLite gateway with vector memory for secure, error-correcting LLM database access.
$29
Aegis-SQL-Bridge
Example session with this skill installed
Calculate total revenue by month for the 'orders' table using the healing bridge for agent 'finance_bot'.
- Read your context and instructions
- Compiled the aegis-sql-bridge
I have executed the query to calculate monthly revenue. The initial attempt failed due to an unsupported PostgreSQL function, but the bridge recalled a self-healing rule to use SQLite's strftime function for the correction.
Monthly Revenue Analysis
| month | total_revenue |
|---|---|
| 2023-01 | 45200.50 |
| 2023-02 | 51300.00 |
| 2023-03 | 48900.75 |
| 2023-04 | 62100.25 |
| 2023-05 | 55400.00 |
Metadata
- Security Status: PASSED
- Execution Time: 2.45ms
- Memory Recall: Rule
9f2a-sqlite-dateapplied (Utility: 1.1) - Token Savings: ~310 tokens via smart schema recall
Next steps
- Call
commit_sql_patternto save this monthly aggregation as a verified hint for 'finance_bot'. - Run
get_bridge_telemetryto check current cache hit rates and token savings. - Use
inspect_smart_schemaif you need to join this revenue data against thecustomerstable.
Connects securely to your tools. The creator never sees your data.
About this skill
Interfacing LLMs with live databases often leads to broken syntax, infinite retry loops, or accidental PII exposure. This skill provides a deterministic, read-only SQLite gateway that uses vector memory to self-heal from SQL errors and enforce zero-trust security.
What it does
- Self-Healing SQL Execution captures SQL errors, distills them into procedural rules using Qdrant, and injects those rules into future attempts to prevent repeated failures.
- Zero-Trust Security enforces a strict read-only driver lock (
?mode=ro) and blocks mutation keywords likeDROP,INSERT, orUPDATEvia AST validation. - Automatic PII Redaction scrubs sensitive data including passwords, SSNs, API keys, and emails from query results before they reach the LLM context.
- Anti-Loop Circuit Breaker uses SHA-256 fingerprinting to detect and halt infinite retry loops for the same failing query pattern.
- Smart Schema Inspection provides token-optimized table metadata, row estimates, and verified SQL hints to help agents write correct joins.
How it works
- Inspect: Call
inspect_smart_schemato retrieve verified hints and table structures without wasting tokens on full DDL. - Execute: Submit queries via
execute_healing_querywhich automatically checks memory for previous failure patterns. - Commit: Save successful, high-performance query patterns to procedural memory using
commit_sql_patternfor future reuse. - Feedback: Update the utility score of recalled rules via
report_rule_outcometo prune ineffective advice.
Frameworks & tools
This skill utilizes SQLite for data persistence, Qdrant for episodic and procedural memory recall, and SHA-256 hashing for its circuit breaker logic. It connects via the MCP (Model Context Protocol) using a Python-based server.
Why this beats prompting it yourself
Prompting an LLM to "be careful" with SQL does not prevent execution timeouts or infinite loops. This tool moves those guardrails into the transport layer, ensuring the agent literally cannot execute mutations and is forced to learn from its own syntax errors.
Use cases
- Safe Data Exploration allowing agents to query production-clones without risk of data loss.
- Automated Analytics where agents must handle dialect-specific SQLite functions like
strftime. - Sensitive Auditing where PII must be masked while still allowing row-level analysis.
- Self-Correcting ETL pipelines that improve their own query logic over time.
Known limitations
Supports single-statement read-only queries only. It cannot execute multiple statements separated by semicolons or perform write operations. Maximum row returns are capped at 100 to prevent context overflow.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Data & Databases.
Supports read-only SELECT/WITH statements only. 100 row hard limit per query. Requires local or in-memory Qdrant instance. SQLite dialect only(EXPLAIN QUERY PLAN is also supported.).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 days ago
- Accounts
Sqlite
Qdrant
- Software
Python 3.11+
- Optional
- SSE / HTTP transport mode (uses local stdio by default).
Enforces strict read-only SQLite access (?mode=ro) via AST/Regex validation and automatic PII/secret redaction. Runs 100% locally in-memory by default with zero external cloud calls unless SSE transport or a remote Qdrant URL is explicitly enabled.
Permissions
Allowed Hosts
File Scopes
Read Files is restricted strictly to read-only SQLite connections (URI ?mode=ro) guarded by an AST/Regex allow-list (SELECT/WITH/EXPLAIN only). Environment Variables configure DB path, row/byte caps, and Qdrant memory settings. Network access is disabled by default unless optional HTTP/SSE transport is enabled.