Codex
Install the plugin
codex plugin marketplace add lyra-forge/marketplace
codex plugin add yoinker@lyra-forge
Invoke it as $yoinker:yoinker
An evidence-first Agent Skill
Give your agent a repository, paper, product, technique, or open question. Get back what it actually does, what the evidence supports, and what—if anything—fits your situation.
A worked example
This compact example uses a real public artifact and a hypothetical local situation. It demonstrates Yoinker’s decision shape; it is not a claim that this adoption happened or improved an outcome.
Would this help us? We want the smallest useful addition, not another architecture platform.
Bottom line
Yes—for executable dependency boundaries in CI. No—as a new source of architectural truth. Start with one rule tied to a failure you already care about.
The rule contract: named from/to constraints, comments that
preserve intent, and error severity that can fail a build.
Add one package-boundary rule against a known regression fixture. Compare signal and maintenance cost with the current lint-only baseline.
Resolution accuracy and runtime on the full monorepo. Expand only if the bounded rule survives real aliases, generated code, and CI timing.
Starting with diagrams or treating the dependency graph as the canonical architecture model. Neither addresses the named regression first.
Install
Install from the Lyra Forge marketplace and start a new session so your harness loads the skill. There is no service, account, API key, or package dependency.
Codex
codex plugin marketplace add lyra-forge/marketplace
codex plugin add yoinker@lyra-forge
Invoke it as $yoinker:yoinker
Claude Code
/plugin marketplace add lyra-forge/marketplace
/plugin install yoinker@lyra-forge
Invoke it as /yoinker:yoinker
Prefer to inspect before installing? Read the entire skill or browse the repository .
How it works
01
“Here is a repository, paper, product, or practice. Does it help this situation, and how should we think about adoption?”
02
“What does this actually do? How do these approaches differ? What does the evidence support, and what remains uncertain?”
Follow discovery links to original artifacts. Inspect the named local context before claiming fit.
Keep observation, source claims, inference, unknowns, recommendations, and outcomes from blurring together.
Judge each mechanism as borrow, try, track, or reject. Name the evidence and what would reverse the call.
Good first prompts
Release status
Version 0.1.0 is the initial MIT-licensed release. Repository checks, both plugin manifests, marketplace installation, and the public deployment are verified.
The held-out behavior comparison is still planned. Availability and a polished example are not presented as evidence that the skill improves outcomes.
Read the evaluation contract