# Relay10 Skill Lab

> Create or tune a focused DisciplinedRun skill using clean-room prompts, trigger tests, and with-skill versus baseline evaluation. Use when adding a skill or changing its description or workflow; do not use merely to run an existing skill.

- **Type:** Skill
- **Install:** `agentstack add skill-minwoo19930301-relay10-relay10-skill-lab`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [minwoo19930301](https://agentstack.voostack.com/s/minwoo19930301)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [minwoo19930301](https://github.com/minwoo19930301)
- **Source:** https://github.com/minwoo19930301/relay10/tree/main/plugins/relay10/skills/relay10-skill-lab

## Install

```sh
agentstack add skill-minwoo19930301-relay10-relay10-skill-lab
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# DisciplinedRun Skill Lab

Add a skill only when it improves a repeated job enough to justify context and maintenance cost.

## Design

1. Define one job, its expected output, and its non-goals.
2. Write a front-loaded description covering what it does, when it triggers, and an important non-trigger boundary.
3. Keep `SKILL.md` short. Move only substantial optional detail into directly linked `references/`; add deterministic scripts only for work the model should not improvise.
4. Write all prompt text and scripts independently. Record third-party behavior patterns and licenses in provenance instead of copying text.
5. Avoid absolute user paths, hidden network calls, interactive scripts, and unbounded retry loops.

## Evaluate

Create at least:

- ten should-trigger prompts;
- ten near-miss should-not-trigger prompts;
- three representative outcome tasks;
- one adversarial or failure case.

Split trigger examples into tuning and validation sets. Compare `with_skill` against `without_skill` on output quality, correctness, token use, time, and failure type. Forward-test in a clean task or subagent so the authoring context does not leak into the result.

Recommended promotion gate: validation recall at least 90%, near-miss false triggers at most 10%, all deterministic assertions passing, and no material quality regression.

## Validate the pack

Check:

- valid YAML frontmatter and matching folder/name;
- concise `agents/openai.yaml` with a literal `$skill-name` starter prompt when the pack targets Codex distribution; Claude Code ignores this file;
- no TODOs, broken relative links, or machine-specific absolute paths;
- deterministic scripts support `--help`, non-interactive execution, clear exit codes, and structured output;
- plugin manifest points only to files that exist;
- total metadata remains small enough for progressive disclosure.

Reject a skill whose baseline is already as good or whose maintenance cost outweighs its measured benefit.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [minwoo19930301](https://github.com/minwoo19930301)
- **Source:** [minwoo19930301/relay10](https://github.com/minwoo19930301/relay10)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-minwoo19930301-relay10-relay10-skill-lab
- Seller: https://agentstack.voostack.com/s/minwoo19930301
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.