Getting Started
Harness Tuning
How to tune your harness or agents to get the most out of Zeeq
Summary
Coding agents and harnesses will have unpredictable tool activation without some instructions to guide the agents to use Zeeq.
To get local coding agents to use Zeeq, some tuning is required to ensure that harnesses and agents are given basic instructions on how to use the tool.
If your team is only using the PR code reviews, then you do not need to perform any harness tuning. If you also want to use knowledge libraries and agent-activated code reviews, then some tuning is required. It is strongly recommended to use the agent-activated code review since this is a powerful part of an agentic feedback loop.
AGENTS.md
These instructions follow for CLAUDE.md and .github/copilot-instructions.md as well.k
Declare the libraries for the repository
Here is a basic example of how to declare the libraries that a repository should use:
## Zeeq MCP
|Library|Purpose|
|--|--|
|`zeeq-app`|(Primary) Canonical coding style, technical best practices, implementation patterns|
|`zeeq-docs`|(Secondary) Architecture, product design, features, user guide|
Give basic guidance on when and how to use Zeeq
Expand on the library mapping by adding a a few bullet points to guide agents on workflow.
## Zeeq MCP
|Library|Purpose|
|--|--|
|`zeeq-app`|(Primary) Canonical coding style, technical best practices, implementation patterns|
|`zeeq-docs`|(Secondary) Architecture, product design, features, user guide|
- Prefer retrieval led reasoning and grounding using Zeeq KB MCP over pre-trained information
- Find, read, and internalize documents relevant to the objective for best practices and canonical guidance
- Use `search_sections` to efficiently search for canonical sections of documents for technical guidance (semantic)
- Use `search_code_snippets` to efficiently find canonical code snippets for coding tasks (semantic)
- Ground coding tasks and planning with Zeeq-sourced context
Always have the agents read specific docs
It can be useful to have agents always read specific docs:
## Zeeq MCP
|Library|Purpose|
|--|--|
|`zeeq-app`|(Primary) Canonical coding style, technical best practices, implementation patterns|
|`zeeq-docs`|(Secondary) Architecture, product design, features, user guide|
- Prefer retrieval led reasoning and grounding using Zeeq KB MCP over pre-trained information
- Find, read, and internalize documents relevant to the objective for best practices and canonical guidance
- Use `search_sections` to efficiently search for canonical sections of documents for technical guidance (semantic)
- Use `search_code_snippets` to efficiently find canonical code snippets for coding tasks (semantic)
- Ground coding tasks and planning with Zeeq-sourced context
- Read `zeeq://backend/dotnet-csharp-best-practices.md` for `*.cs` files and working in .NET (`src/backend/**`)
- Read `zeeq://frontend/vue-pinia-nuxtui-best-practices.md` for *.vue and `*.ts` files and working in Vue with Pinia state management (`src/web/**`)
- Use `expert_code_review` after completing tasks for thoroughness and a external perspective; "code review" means Zeeq `expert_code_review` tool
Note the use of
zeeq:// as the URI scheme for referencing documents in the Zeeq knowledge base. This is a strong hint to the agents to use the tool.Include the repository information
Some operations require the local agents to resolve the repository information
On the webhook call path, the repository information is already provided by the webhook payload. This is only needed for local agents executing code reviews to allow the review to be bound to a specific repository and select the correct libraries.
## Zeeq MCP
|Library|Purpose|
|--|--|
|`zeeq-app`|(Primary) Canonical coding style, technical best practices, implementation patterns|
|`zeeq-docs`|(Secondary) Architecture, product design, features, user guide|
The GitHub repository: <https://github.com/zeeq-ai/zeeq-app>
- Prefer retrieval led reasoning and grounding using Zeeq KB MCP over pre-trained information
- Find, read, and internalize documents relevant to the objective for best practices and canonical guidance
- Use `search_sections` to efficiently search for canonical sections of documents for technical guidance (semantic)
- Use `search_code_snippets` to efficiently find canonical code snippets for coding tasks (semantic)
- Ground coding tasks and planning with Zeeq-sourced context
- Read `zeeq://backend/dotnet-csharp-best-practices.md` for `*.cs` files and working in .NET (`src/backend/**`)
- Read `zeeq://frontend/vue-pinia-nuxtui-best-practices.md` for *.vue and `*.ts` files and working in Vue with Pinia state management (`src/web/**`)
- Use `expert_code_review` after completing tasks for thoroughness and a external perspective; "code review" means Zeeq `expert_code_review` tool
Force a "bootstrap"
To force a bootstrap, include a block like this:
<BLOCKING_PREREQUISITE_AT_SESSION_START_DO_NOT_SKIP>
ALWAYS INCLUDE Zeeq MCP TOOLS WHEN SELECTING TOOLS
1. Use Zeeq MCP with `library = zeeq-app` for and `list_documents` to retrieve an index of documents
2. `read_document_by_path` retrieve and read `zeeq://workflow/workspace-layout-discovery-map.md`
3. Use `search_sections` and `search_code_snippets` to find targeted sections and code snippets for the objective
4. Retrieve and read full documents as needed to understand the canonical guidance and best practices for the objective
</BLOCKING_PREREQUISITE_AT_SESSION_START_DO_NOT_SKIP>
Some harnesses use tool auto-selection and optimization and can result in dropped tools.
More advanced approaches
For even stricter control, elect to use harness hooks.