Issues and PRs
Discuss usage and improve skill code in the repository.
Lowest additional setup.
Experience is less structured for
recall.
01 / The building blocks
What a skill contributes
- Skill ↗
- Task-specific instructions and resources, loaded when relevant.
- MCP server
- A program exposing tools, resources, and prompts through a standard protocol.
- Connector ↗
- An integration with an external service. Terminology varies by product.
- Hook ↗
- Configured behavior at an agent host’s lifecycle event.
These components can work together. A skill can explain how to use connected tools.
02 / Existing efforts
Workshop’s place in the landscape
| Existing effort | What it provides | Workshop’s relationship |
|---|---|---|
|
ASM,
Vercel Skills, gh skill ↗ |
Discovery and skill management | Delegates discovery |
| APM ↗ | Reproducible agent dependencies | Delegates external dependencies |
| SkillPort ↗ | Skill access across agents through CLI or MCP | Overlaps on portable access |
| SkillNote ↗ | Registry, synchronization, and agent reviews | Closest overlap on feedback |
Workshop’s focus
Coordinate these tools and retain decisions and experience across projects in Git.
Selected landscape, checked 18 September 2026. Multi-agent support and feedback already exist elsewhere.
03 / Experience so far
More skills are entering my work
Discoverability improved,
and I started using skills more.John’s experience, paraphrased
25remembered skills
5projects
37use observations
6contribution records
Vendor-agnostic design. Ordinary skill files, portable memory, replaceable integrations.
A reversible commitment. Projects retain their skills and dependency state if Workshop goes away.
Memory snapshot ↗ · 18 September 2026 · Activity counts, not measured efficiency gains. No controlled evaluations recorded.
04 / Live demo · 6–8 minutes
A task, a skill, and what we learned
Use a real con/skills licensing workflow ↗ throughout.
-
1
Recall earlier work
Recover a skill from the task, without knowing its name.
pixi run workshop recall "license compliance" -
2
Compare sources
See local candidates and public discovery in one workflow.
pixi run workshop find "license compliance" -
3
Inspect the experience
Read what helped and what needed correction before choosing.
pixi run workshop history introduce-reuse-compliance -
4
Close the loop
Inspect actual feedback and proposed improvements.
pixi run workshop insights
Discovery now includes K-Dense, NiPreps, con/skills, Anthropic, OpenAI, and configured installed skill roots.
05 / The next usage cycle
Feedback after meaningful use
Actual observation · condensed
introduce-reuse-compliance
- Task
- Add REUSE licensing checks to a project.
- Helped
- Guided licensing structure, validation, and pre-commit integration.
- Correction
- The pre-commit hook needed a project-local wrapper for Python compatibility.
Source observation ↗ · agent-reported, unreviewed
Ask agents to report
what actually changed.
Keep failures and unclear benefit.
Make ratings optional.
Use recall and insights to inform the next choice or
skill improvement.
Current trigger: global instructions plus
workshop-feedback ↗.
No lifecycle hook ships yet. Capture can be missed.
Next trial: does feedback change later choices, expose failures, or improve a skill without much bookkeeping?
06 / Proposal for con/skills
Shared skills, shared experience
Keep developing skills in con/skills ↗. Compare ways to share what happens when we use them.
SkillNote ↗
Use a shared service for skills, synchronization, and reviews.
Richer interface.
Assess hosting and history export.
Workshop memory
Contribute sanitized observations through Git. Recall colleagues’ experience.
Portable records.
Group workflow still needs testing.
A few members. Two recurring tasks. Four weeks.
Keep agent reports attributed. Review shared records. Look for useful reuse and improvements to con/skills.
Choose the lightest approach that helps the group. APM or other distribution tools can serve any option.