Introduction¶
When using coding agents, we often encounter this problem: the model is capable, but the execution process lacks discipline. It may edit production code too early, lose the original plan, rely on outdated API knowledge, or claim “Done” without sufficient validation. This is usually not the model’s fault, but the absence of a well-designed “harness.”
get-fable is a tool designed to provide this discipline for coding agents. It does not replace the underlying model; instead, it builds a deterministic workflow and skill routing around the model, turning “good work habits” into executable steps.
What Is It¶
get-fable is a portable coding lifecycle system. It wraps the agent in a structured workflow and guides decisions through 13 core skill modules. It is maintained by imMamdouhaboammar and uses the MIT license.
Its core goal is to shift the agent from “guessing” to “evidence-driven” operation, ensuring each action has a corresponding validation step instead of relying solely on the model’s raw reasoning ability.
Core Features¶
get-fable provides the following capabilities:
- End-to-end lifecycle: Covers the complete closed loop from discovery, research, planning, testing, delegation, execution, validation, review, security hardening, recovery, to release.
- Deterministic routing: Based on task type and context, intelligently selects the next dedicated skill module instead of using the same set of instructions for all requests.
- Core skills: Includes routing, research, planning, architecture design, TDD (test-driven development), validation, review, red-teaming, security healing, recovery, gRPC Worker, and TOON protocol support.
- DSH plugin integration: As a DeepSeek Harness plugin, it works seamlessly with the DSH environment.
Installation and Enablement¶
Install it globally using the Bun package manager:
bun add -g get-fable
After installation, it runs as a DSH plugin and takes over workflow control for the current agent.
Typical Usage¶
get-fable defines a standardized operation sequence. For a typical bug-fix task, it enforces the following steps:
- DISCOVER: Trace the actual codebase/runtime execution paths instead of guessing based on file names alone.
- RESEARCH: Confirm that the current external facts align with the intended version.
- PLAN: Generate evidence-based, dependency-aware, risk-aware, and falsifiable work cards.
- TEST: Follow the TDD process: start with a failing test (RED), then modify the code to make the test pass (GREEN), and verify the mutation testing results.
- REVIEW: Review the actual code changes.
The full lifecycle can be summarized as:
DISCOVER -> RESEARCH -> PLAN -> TEST -> DELEGATE -> EXECUTE -> VERIFY -> REVIEW -> SECURITY -> RECOVER -> RELEASE
For example, when fixing a token refresh race condition, the system guides you to first trace the behavior, reproduce the issue, and confirm the test boundaries, then make minimal production code changes, and finally validate the mutation results and review the diff, instead of modifying code directly.
Use Cases and Notes¶
- Intended audience: Development teams that need to improve the execution discipline of coding agents, or developers who want to move away from “one-size-fits-all” system prompts and toward reproducible workflows.
- Permissions and security: As a DSH plugin, it runs with the permissions of the current DSH process. Before enabling it, it is recommended to review its source code and license (MIT).
- Positioning: It is not a Prompt Pack; instead, it provides “operational judgment around instructions” to improve the stability and reliability of the model.
Summary¶
get-fable demonstrates that an agent’s strength depends not only on the model itself, but also on the way it works around it. By introducing deterministic skill routing and validation steps, it elevates agent behavior from “trial fixes” to the level of “rigorous engineering”.