By drmaciver
Enforce reliability in Claude Code workflows by prioritizing unblocked tasks from P0-P4 backlogs via CLI, achieving 100% Rust test coverage with gap analysis and cargo llvm-cov, diagnosing/fixing GitHub Actions failures, creating structured how-to docs, handling issues with acceptance criteria, gathering requirements, and enhancing tools with research hooks.
npx claudepluginhub drmaciver/claude-reliability --plugin claude-reliabilityThis skill should be used when the user asks "what should I work on", "what's next", "pick a task", or when needing to select between multiple available tasks. Provides guidance for prioritizing and selecting tasks.
This skill should be used when the user asks to "fix the CI", "fix the build", "fix GitHub Actions", "why is the build failing", or mentions a failing CI/CD pipeline. Provides systematic workflow for diagnosing and fixing GitHub Actions failures.
This skill should be used when the user asks to 'get to 100% coverage', 'improve test coverage', 'cover uncovered code', 'fix coverage gaps', or mentions uncovered lines or coverage percentage. Provides strategies for achieving complete test coverage through better code design.
This skill should be used when the user asks 'what should I work on', 'what's next', 'pick a task', or when needing to select between multiple available tasks. Provides guidance for prioritizing and selecting tasks.
This skill should be used when creating documentation, writing instructions, capturing reusable procedures, or when the user asks to 'create a how-to' or 'document a procedure'. Provides guidance for creating effective how-to documentation.
This skill should be used when the user asks to work on a GitHub issue, fix a bug from an issue, or implement a feature request from GitHub. Provides guidance on properly understanding issues, defining acceptance criteria, and committing work that references issues correctly.
This skill should be used when the user asks to implement a new feature, add new functionality, or build something new. Provides a systematic approach to feature implementation that avoids common pitfalls like making false assumptions about APIs or tool capabilities.
This skill should be used when the user's request is ambiguous, the scope is unclear, there are multiple valid interpretations, or details need clarification before proceeding. Guides collaborative requirements gathering to understand the problem being solved.
This skill should be used when starting complex multi-step work, when breaking down work into smaller pieces, or when needing to track progress on tasks. Provides guidance on creating fine-grained tasks with proper dependencies.
This skill should be used before planning or implementing any non-trivial request. It provides a systematic approach to understanding what the user actually wants, not just what their words literally say. Prevents building technically-correct-but-useless implementations by identifying load-bearing features, surfacing obvious implications, and sanity-checking interpretations.
This is a plugin that contains a grab bag of my personal infrastructure for trying to get Claude to be something I can actually delegate real software development tasks to. It's idiosyncratic, a work in progress, only sometimes effective, and every time I try to use Claude without it now I get sad.
In theory if you install it, Claude will become magically better at software development, at the cost of potentially taking quite a lot longer (mostly, but not always, because it actually does what you ask it to do rather than stopping 10% of the way through).
I do not necessarily expect it to be useful to other people in its current state, but if you want to give it a go, I'm interested in hearing other people's experiences with it.
Broadly speaking the problems I am trying to solve are that Claude doesn't do what it's told and what it does doesn't work. This plugin tries to guide it away from common shortcuts it tries to take, and to enforce that it actually does the work that it is told to do. Claude really doesn't like doing that, so it's at best a partial success, but it's a lot better than not using it.
It is particularly optimised for a problem I keep running into where I ask Claude to do something that will take a couple of hours, go away, and come back later to find that 20 minutes in it decided it had done enough work and stopped, or that the task that I had asked it to do was too hard and so it should implement a simpler and obviously useless version of it.
Key features:
There are a number of escape hatches and special cases in here to try to keep Claude also useful for normal interaction while this is going on. They even mostly work.
The plugin also provides a number of skills around general software development and how to do it properly. I'm not actually sure how useful they are, they're a bit speculative.
Step 1: Add the marketplace
/plugin marketplace add DRMacIver/claude-reliability
Step 2: Install the plugin
/plugin install claude-reliability@claude-reliability-marketplace
The plugin will automatically download the pre-built binary from the latest GitHub release, or build from source if no release is available for your platform. You will need a rust compiler if you're building it from source.
| Platform | Method |
|---|---|
| Linux x86_64 | Pre-built release |
| Linux ARM64 | Pre-built release |
| macOS ARM64 | Pre-built release |
| Other Unix Platforms | Builds from source (requires or will install Rust) |
It probably could be made to work on Windows if you wanted it to, but it currently relies on bash scripts that would need a Windows equivalent. I imagine it works fine on WSL but I haven't tested it.
This project is largely vibe-coded, with Claude doing the majority of the development (with mostly-careful oversight from me). In theory the following command will drop you into a claude code shell in a docker container with all of the reliability features present:
just develop
In practice, this command has only ever been tested on my computer and CI, so you'll probably run into issues. Again, please let me know if you do.
MIT
Prompt engineering techniques for accurate, grounded Claude responses — anti-hallucination workflow with citation-backed analysis
Share bugs, ideas, or general feedback.
Opinionated workflow guides and best practices - the preacher's proven patterns for Claude Code projects
SDLC enforcement for AI agents — TDD, planning, self-review, CI shepherd
Verify task completion before finishing work. Catches incomplete work, bugs, and quality issues.
Language-agnostic development process harness implementing the Stateless Agent Methodology (SAM) 7-stage pipeline with ARL human touchpoint model and Voltron-style language plugin composition. Provides orchestration, workflows, planning, verification, and testing methodology that any language plugin can compose with.
Comprehensive skills library for Claude Code: planning, design, TDD, debugging, collaboration patterns, and proven techniques