eval-writer
v1.3.0by tiennguyentt · author: Tien Nguyen
Defines measurable success criteria and designs evaluations for any LLM-based task, product, feature, or piece of logic.
What eval-writer does, in the publisher's words
One independent agent plugin in the agent-plugins marketplace:
This is one of several agent plugins in this repository, where agent teams do the work end to end. It is published on its own because a part that only runs inside the repo that grew it is not a part, it is a dependency.
Built for a team, not for a demo. One skill with one job, invocable without the agent. Its grading standard is a rubric.md file read by an agent that did not produce the work, which is exactly what makes a cheaper, independent verifier viable. The packaging is checked by machine: plugin.json conforms to Agent Plugins 1.0.0, and a contract check fails the build when the manifests, the entry skill, or the routes drift apart.
From the project README, punctuation lightly normalized · Open the README on GitHub
- Skills
- 1
- MCP servers
- 0
- Stars
- 0
- License
- MIT
- Repo created
- 2026-07-31
- Last pushed
- 2026-08-14
- Publisher type
- User
- Version
- 1.3.0
Links
Skills · 1
- write-success-criteria
- Defines measurable success criteria and designs an evaluation for any LLM-based task, product, feature, or piece of logic, not limited to capabilities. Use when the user asks "how do I know if this is working", "what does success look like for X", "help me design an eval for this feature", "cần ti…
Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.
Category
Agent Tooling. Assigned by this directory. Agent Plugins 1.0.0 has no category field, so no manifest declares one.
Keywords
personal-os · eval-design · success-criteria
Related by publisher and keywords
Creates, evaluates, and packages confirmed capabilities as one-agent-with-licensed-skills plugins.
Implements features scenario-first: Given/When/Then scenarios seen failing, the three laws of TDD, and a measured quality gate before every commit.
Tests a macOS app's UI hands-free: off-screen launch, Accessibility-tree gates, screenshot capture with vision review, and golden AX-tree snapshot regression, every number measured, nothing advertised it cannot deliver.
Use Claude Code from Codex to review code or delegate tasks.
Classifies the unknown, picks the artifact form, and writes one planning document at a time from the five-document chain.
Open-source education plugin with formative assessment and student feedback skills for AI agents: feedback that moves learners forward, assessment design, hinge questions, success criteria, peer and self-assessment, and responsive analysis of student work.
[](https://agentpluginsdirectory.com/plugins/eval-writer)✓ Verified . We fetched the manifest from GitHub and checked it against the official Agent Plugins 1.0.0 schema at agent-plugins.org.