Design partners wanted: UAT the Skillsmith Agent on your coding harness
We just shipped the Skillsmith Agent (@skillsmith/mcp-server 0.7.0+, sklx 0.8.0+): a portable agent that keeps your installed skills current, audits your local inventory for collisions, and vets a skill before you install it. It runs inside your coding agent, not as a separate tool you have to remember to invoke.
Before we open this up broadly, we want a small group of testers on each harness to run a scripted checklist and tell us what breaks or feels off. This issue is the sign-up sheet.
What testing looks like
- Install with one command (
sklx agent install).
- Run three jobs: check for outdated skills, audit your inventory for namespace collisions, and vet a new skill before installing it.
- Try the undo path, the session-start nudge, and (where supported) the upgrade prompt.
- Uninstall. It reverses everything it wrote.
Each harness has a slightly different script and takes 45 to 85 minutes. Full breakdown below.
| Harness |
Support tier |
What's specific to it |
Est. time |
| Claude Code |
1 |
/skillsmith invocation, session-start nudge |
~75 min |
| Cursor |
1 |
.claude/agents shim, hooks.json nudge, tool-budget sanity |
~80 min |
| Codex / OpenAI |
1 |
TOML agent config, .agents/skills load, approval flow |
~70 min |
| Copilot / VS Code |
1 |
Both VS Code agent mode and CLI (.agent.md), tool-cap with a second MCP server running |
~85 min |
| OpenCode |
2 (best-effort) |
Skill discovery and basic MCP invocation |
~45 min |
| Hermes |
2 (best-effort) |
Skill load + MCP tools, no session hooks on this platform |
~60 min |
| Windsurf / Devin |
3 (skill-only) |
Auto-trigger and the 100-tool shared budget |
~50 min |
How to sign up
Comment below with the harness(es) you use regularly. We'll follow up with the install one-liner, the checklist script, and a feedback form. We're prioritizing Tier 1 harnesses first, but we want signal from every tier.
You do not need a paid plan. Community tier quota is more than enough for the script.
FAQ
Will you collect data about my use?
Yes, but only what you opt into. A consent-gated telemetry layer records which tools the agent calls, how long they take, and whether they succeed. We ask you to opt in from your account dashboard as part of the pilot, since the pilot's success measure depends on those events. We never collect your code, your prompts, or the contents of your skills.
How is my identity recorded?
We assign you a partner code for tracking and feedback. Your name is never stored against test results.
What if something breaks or I want to back out?
Uninstall with one command at any point, before, during, or after the test. It leaves no trace.
What do I do if I hit a problem during testing?
Reply on this issue or in the channel we send you with the sign-up confirmation. We aim to respond within 24 hours on weekdays.
Thanks for helping us get this right before it ships broadly.
Design partners wanted: UAT the Skillsmith Agent on your coding harness
We just shipped the Skillsmith Agent (
@skillsmith/mcp-server0.7.0+,sklx0.8.0+): a portable agent that keeps your installed skills current, audits your local inventory for collisions, and vets a skill before you install it. It runs inside your coding agent, not as a separate tool you have to remember to invoke.Before we open this up broadly, we want a small group of testers on each harness to run a scripted checklist and tell us what breaks or feels off. This issue is the sign-up sheet.
What testing looks like
sklx agent install).Each harness has a slightly different script and takes 45 to 85 minutes. Full breakdown below.
/skillsmithinvocation, session-start nudge.claude/agentsshim, hooks.json nudge, tool-budget sanity.agents/skillsload, approval flow.agent.md), tool-cap with a second MCP server runningHow to sign up
Comment below with the harness(es) you use regularly. We'll follow up with the install one-liner, the checklist script, and a feedback form. We're prioritizing Tier 1 harnesses first, but we want signal from every tier.
You do not need a paid plan. Community tier quota is more than enough for the script.
FAQ
Will you collect data about my use?
Yes, but only what you opt into. A consent-gated telemetry layer records which tools the agent calls, how long they take, and whether they succeed. We ask you to opt in from your account dashboard as part of the pilot, since the pilot's success measure depends on those events. We never collect your code, your prompts, or the contents of your skills.
How is my identity recorded?
We assign you a partner code for tracking and feedback. Your name is never stored against test results.
What if something breaks or I want to back out?
Uninstall with one command at any point, before, during, or after the test. It leaves no trace.
What do I do if I hit a problem during testing?
Reply on this issue or in the channel we send you with the sign-up confirmation. We aim to respond within 24 hours on weekdays.
Thanks for helping us get this right before it ships broadly.