Segment 1: HiddenMerit Morning Post · Issue 86

Focus on Database Frontiers, Practical Insights for DBAs August 26, 2026 ｜ 5 Selected Global Breaking News

01｜OceanBase Series A Funding in Progress: Targeting RMB 2‑3 Billion, Annualized Revenue Exceeds RMB 1.4 Billion, Benchmarking Databricks

Multiple financial media outlets have confirmed that Ant Group’s independent database company OceanBase is progressing with its Series A funding round, targeting approximately RMB 2‑3 billion, and has already engaged with multiple leading investment institutions. This marks OceanBase’s first external funding since its establishment.

Key Data: In 2026, OceanBase’s annualized revenue has exceeded RMB 1.4 billion (approximately $200 million) , up approximately 70% year‑on‑year. The company has surpassed 4,000 global customers, with the non‑financial customer ratio expected to approach 50‑50 in 2026. According to IDC data, OceanBase ranked first in China’s distributed database market in 2025.

Strategic Benchmarking: OceanBase has been referred to in the media as “China’s Databricks” – starting from distributed databases and expanding into enterprise‑wide data management, agent data foundations, and data services. OceanBase CEO Yang Bing previously stated that the company aims to go public in approximately three to four years.

• DBA Perspective: OceanBase’s first external funding is a landmark event for domestic databases moving toward “market‑oriented independent operations.” Benchmarking against Databricks means its strategic direction is extending from a “database company” to an “AI data platform” – DBA skills need to expand from “distributed database operations” to “AI data platform architecture and multi‑modal data management.”

02｜GitHub Trend: Agent Skill Ecosystem Enters “Consolidation Phase,” deepseek‑harness Tops with 3,334 New Stars in a Single Day

On August 25, the GitHub Trending榜单 saw a significant shift. deepseek‑ai/deepseek‑harness topped the chart for the first time with 3,334 new stars in a single day. The project has surpassed 175,000 stars in less than two weeks since launch – a rare growth rate in GitHub history.

deepseek‑harness is an AI Agent runtime framework built on the design philosophy of “everything is pluggable,” covering models, tools, policies, storage, and context management – all replaceable. The official release already includes over 100 plugins, with a community plugin ecosystem rapidly forming. The “Android of the Agent era” is taking shape.

The Skill Ecosystem Enters the “Consolidation Phase”: A few days ago, mattpocock’s personal .agents directory was trending; today it has dropped off the list, replaced by aggregators and curated collections. VoltAgent, with 1,497 Skills, gained 600 stars in a single day. Andrej Karpathy’s single‑file CLAUDE.md with four rules has reached 206,000 stars total. Individual workshops are receding; aggregators and curated selections are taking the stage – in just three days, the shift from “everyone putting products on the shelf” to “who becomes the shelf” has unfolded.

📌 HiddenMerit Mid‑Day Preview: Today at noon, we will dive deep into deepseek‑harness’s database plugin ecosystem – when agents can directly read databases, write SQL, and query schemas, how will DBA operations change?

• DBA Perspective: The “aggregation” trend in the Agent Skill ecosystem means AI Agent capabilities are moving from a “wild ecosystem” to “standardization.” For DBAs, standardized Skill encapsulation means database operations knowledge can be reused and distributed through Agent Skills – the core value of DBAs will shift from “writing scripts” to “沉淀 reusable operational Skills.”

03｜Databricks Acquires Electric: Bringing Local Postgres Database to AI Agents

Recently, Databricks announced the acquisition of startup Electric, which embeds Postgres databases into application environments using WebAssembly technology. Electric’s PGLite, combined with the Electric Sync real‑time sync engine, allows agents to run Postgres‑compatible databases locally while staying synchronised with the central database. After the acquisition, PGLite will form a two‑tier database architecture with Databricks’ Lakebase – PGLite handles agent local data processing, while Lakebase serves as the centralised database for shared persistent data.

Analyst Views: The lead analyst at Pareekh Consulting noted that local database access reduces network hops, helping improve agent speed, especially for complex, long‑running tasks. The lead analyst at Nord‑IQ Research cautioned about data governance and security concerns – distributing state across agent sandboxes expands the attack surface, forcing enterprises to extend access control, auditing, and compliance frameworks from a single database boundary to hundreds of temporary local instances.

• DBA Perspective: Databricks’ acquisition of Electric reflects the urgent need for “data locality” in AI agent applications. For DBAs, database architecture is evolving from a “single central database” to a “central + local two‑tier architecture.” Sandbox‑level data governance and security challenges remind DBAs that AI agent data management cannot follow traditional models – new frameworks for state data location, retention, auditing, and cleanup need to be established.

04｜CVE-2026-14669 (CVSS 8.8): PostgreSQL to_char() Heap Overflow PoC Publicly Released

The PostgreSQL heap buffer overflow vulnerability (CVE-2026-14669, CVSS 8.8) disclosed on August 13 now has full technical details and PoC publicly released, significantly increasing the risk. The vulnerability stems from the to_char() function failing to perform length checks when processing overly long POSIX time zone abbreviations – attackers can exploit this with an overly long time zone abbreviation to trigger a heap overflow and achieve code execution.

Affected Versions: All PostgreSQL versions before 18.5, 17.11, 16.15, 15.19, and 14.24. Given PostgreSQL’s widespread deployment as the backend for countless web applications and services, the impact scope is enormous.

Remediation: The PostgreSQL team fixed this vulnerability in the August 2026 update. With the PoC now publicly available, all affected users should upgrade immediately and not rely solely on network‑layer protections.

• DBA Perspective: This is another high‑risk PostgreSQL vulnerability following CVE-2026-6472 and CVE-2026-6473 this year. The CVSS 8.8 rating combined with a public PoC means the exploitation barrier has been significantly lowered. to_char() is a widely used time‑formatting function – any user with SQL execution privileges can trigger it. DBAs are advised to immediately check versions and prioritise upgrades within the Q3 patch window.

05｜PostgreSQL 14 EOL Countdown: Only 78 Days Until End of Support

PostgreSQL 14 will officially stop receiving patch updates on November 12, 2026 – only approximately 78 days remain.

Recommendation: Enterprises still running PG 14 in production environments must immediately formulate upgrade plans and migrate to supported versions such as PG 17 or 18. PG 14 released its final version 14.23 on May 14, with no further security patches or bug fixes. PG 18 is the current recommended LTS version, with support extending to May 2031.

• DBA Perspective: With less than 3 months until PG 14 EOL, DBAs must immediately initiate upgrade assessments. It is recommended to use pg_upgrade --check for compatibility pre‑checks, with special attention to extension module dependencies and postgresql.conf parameter differences.

📌 Mid‑Day & Night Talk Previews

📌 HiddenMerit Mid‑Day (Today at Noon): When Agents Can Query Databases – Deep Dive into deepseek‑harness’s Database Plugin Ecosystem

• From dsh‑sql to db‑connector: what does the security model look like for agents directly connecting to databases?
• How can DBA operational knowledge be沉淀 into Agent Skills?
• Long‑term impact on the operations role: from “typing commands” to “defining Skills”

📌 HiddenMerit Night Talk (Tonight): The “Repetitive Work” Stolen by AI Agents – Your Operations Experience Is Being Encapsulated into Skills

• When DatabaseClaw and deepseek‑harness can automatically execute 90% of daily operations, what remains for DBAs?
• From “executor” to “Skill definer”: three pathways for DBA capability evolution
• Night Talk interaction: would you be willing to encapsulate your “hard‑earned” troubleshooting experience into Skills?

HiddenMerit Team Production Slogan: 绩优隐于内，金石启新程 | Hidden deep. Merit bold. Forge ahead.

Segment 2: HiddenMerit Mid‑Day · Issue 3

Focus on Database Tools Ecosystem August 26, 2026 ｜ Deep Dive into 1 Hot Tool

This Issue’s Tool: deepseek‑harness’s Database Plugin Ecosystem

When agents can directly query databases, the way DBAs operate is being rewritten.

1. What is deepseek‑harness?

On August 13, DeepSeek officially open‑sourced DeepSeek Harness (DSH), an AI Agent runtime framework built on the design philosophy of “everything is pluggable.” The official formula is: Model + Harness = Agent.

DSH designs models, tools, policies, storage, and context management as all replaceable modules, with over 100 plugins already built into the official release. In less than two weeks since launch, it has surpassed 175,000 GitHub Stars.

For DBAs, DSH’s database plugin ecosystem is particularly noteworthy. Agents are evolving from “just chatting” to “being able to query databases themselves.”

2. Database Plugin Ecosystem Panorama

Multiple database‑related plugins have already emerged in the DSH community:

1. dsh‑sql (The Most Active Community Database Plugin)

Maintained by community developers, supports SQLite, MySQL, and PostgreSQL, providing four core tools:

Tool	Function	Security Mechanism
sql_list	List configured connections	Read‑only
sql_query	Execute SELECT queries	Lexical‑level allowlist validation
sql_exec	Execute write operations/DDL	Write approval gate (popup approval by default)
sql_schema	View table structures and column information	Identifier allowlist validation

Security design worth noting: sql_query strips strings and comments before keyword validation, rejecting multi‑statements, SELECT INTO, FOR UPDATE, and other risky operations. readOnly: true can globally disable write operations. Each connection can be independently configured with maxRows to prevent large queries from exhausting resources.

2. db‑connector (Enterprise‑Grade Security Design)

A database connector maintained by community developer JohnXu22786, designed for “read‑only security, write approval gate, JSONL audit trail,” supporting SQLite/PostgreSQL/MySQL. Core security mechanisms include:

• Credentials never appear: configuration only references environment variable names or credential service references – no plain‑text passwords stored
• Audit logs: each SQL operation is automatically recorded to a JSONL file, containing statement summary, row count, and execution time
• Write approval gate: sensitive operations require approval before execution

3. KaleidoSphere for DSH (Enterprise‑Grade Database Analysis Plugin)

A professional database analysis plugin maintained by JoFe2, supporting Microsoft SQL Server and Oracle with native paths – no separate client driver installation required. Provides six tools: status, discovery, analyze, plan, preview, and readback, supporting read‑only analysis workflows.

4. Database‑Related Skills in the Official Skill Ecosystem

The DSH official Skill ecosystem already includes multiple database‑related Skills: dsh-sqlite (SQLite database operations), dsh-db (generic database connection), dsh-supabase (Supabase integration), and dsh-prisma (Prisma ORM integration).

3. What Can DBAs Do with It?

Scenario 1: Natural Language Slow SQL Analysis

Operations personnel describe problems in natural language, and the agent automatically queries the database, locates slow queries, and provides analysis conclusions.

Scenario 2: Schema Visualisation and Documentation Generation

The agent automatically reads metadata and generates ER diagrams or Markdown documentation for technical documentation沉淀.

Scenario 3: Automated Data Validation

The agent reads two databases, compares table structures and data differences, and outputs discrepancy reports.

4. Security Boundaries: Three Principles for Agents Connecting Directly to Databases

1. Least Privilege: configure read‑only accounts for agents, and control changes through write approval gates when necessary
2. Auditability: all SQL operations are recorded in audit logs
3. Row Limiting: limit query return row counts to prevent large queries from overwhelming the database

dsh‑sql’s three‑tier security mechanisms – readOnly, writeApproval, and maxRows – provide a reference defence model for agents connecting directly to databases.

5. Long‑Term Impact on the DBA Role

The maturity of the Agent Skill ecosystem means: DBA operational knowledge is moving from “stored in the mind” to “encapsulated in Skills.” In the future, a DBA’s core competitiveness will shift from “how many commands are memorised” to “how many reusable operational Skills are沉淀.”

This is not replacement – it’s leverage.

HiddenMerit Team Production Slogan: 绩优隐于内，金石启新程 | Hidden deep. Merit bold. Forge ahead.

Segment 3: HiddenMerit Night Talk · Issue 18

DBA Workplace & Life August 26, 2026 ｜ Let’s Talk About DBA Life

Tonight’s Topic: When Your Operational Experience Is Encapsulated into Skills – What’s Left for DBAs?

1. A Trend That Is Happening

deepseek‑harness has been open‑sourced, and the Agent Skill ecosystem is entering its “consolidation phase.”

This is not just another tech news story – it means: the troubleshooting experience you’ve accumulated over a decade is being progressively packaged into Skills that others can also call.

How to check slow SQL, how to locate replication lag, how to restore backups… these “craft skills” that once kept you up at night are becoming standardised operations that agents can execute with a single click.

2. Will DBAs Be Replaced?

Here’s the conclusion: No. But the DBA who “only knows how to do it” will be replaced by the DBA who “knows how to define Skills.”

A DBA’s core capability has never been “executing commands” – it’s three things:

1. Knowing what to look for: being able to locate the root cause of a problem, not just seeing the symptoms
2. Knowing how to decide: choosing the best option for the current scenario among multiple alternatives
3. Knowing how to recover: being able to stop the bleeding quickly when things go wrong

Agents can’t do these three things yet. When an agent doesn’t know which table to query or which index structure is more appropriate, a human must set the direction – and the work agents do happens to be exactly what DBAs consider repetitive labour.

The essence of an Agent Skill is packaging “things that have already been figured out” into reusable modules.

So agents aren’t taking DBAs’ jobs – they’re freeing DBAs from repetitive labour, giving them the bandwidth to focus on higher‑value judgment and design.

3. From “Executor” to “Skill Definer”: Three Pathways for DBA Capability Evolution

Path 1: Building an Operational Skill Library

Slow SQL diagnosis, lock analysis, primary‑standby failover – distil daily repetitive tasks into standardised operating procedures and find opportunities to package them into reusable Skills.

Path 2: From “Managing Databases” to “Managing Data Architecture”

Stop looking at just a single database instance; see the entire enterprise’s data flow. Understand the business, design data models, plan capacity, define standards.

Path 3: From “Execution” to “Design”

A DBA executes after receiving a request – create table, add index, tune parameters. A data architect designs before the request even comes – what data capabilities might the business need in the future? Can it still handle it when data volume grows 10x in three years?

Just as DevOps沉淀 operational knowledge into IaC (Infrastructure as Code), in the AI era, DBAs can沉淀 operational knowledge into Skill as Code. The difference: IaC manages infrastructure, while Skills manage “how to do something.” The former standardises server configuration; the latter makes operational experience reusable.

4. Tonight’s Interaction

Feel free to share your thoughts in the comments:

1. What is the most “repetitive” task in your daily work?
2. If you were to package it into a Skill, how would you design it?
3. What do you think is the biggest barrier for DBAs transitioning to data architects?

HiddenMerit Team Production Slogan: 绩优隐于内，金石启新程 | Hidden deep. Merit bold. Forge ahead.