# Sources and claim notes

Selected landscape reviewed on 18 September 2026.
This is not an exhaustive survey or a claim that Workshop has no competitor.
Project links describe capabilities, not a hands-on benchmark of every tool.

## Slide 1: definitions

- [Agent Skills overview](https://agentskills.io/home) and [specification](https://agentskills.io/specification): a skill directory contains `SKILL.md`, metadata, task instructions, and optional resources.
  The host can expose descriptions and load full instructions when relevant.
- [MCP architecture](https://modelcontextprotocol.io/docs/learn/architecture): servers expose tools, resources, and prompts through MCP.
- [Claude connectors](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities): a concrete example of service integration through connectors.
  Connector terminology and implementation are product-specific.
- [Claude Code hooks](https://code.claude.com/docs/en/hooks): a concrete example of behavior configured for lifecycle events.
  Hook support varies by host.

These mechanisms are complementary, not interchangeable skill formats.
The release-review example is illustrative, not a recorded evaluation.

## Slide 2: landscape

| Project | Supported characterization | Relationship to Workshop |
| --- | --- | --- |
| [ASM](https://github.com/luongnv89/asm) | Cross-provider inventory, search, and management | Workshop delegates discovery, not its full feature set |
| [Vercel Skills](https://github.com/vercel-labs/skills) | Open skill discovery and installation tooling | Workshop delegates discovery |
| [gh skill](https://cli.github.com/manual/gh_skill) | GitHub search, preview, installation, update, and publication | Workshop uses discovery and preview |
| [Microsoft APM](https://github.com/microsoft/apm) | Manifest and lock for reproducible agent dependencies | Owns accepted external project dependencies |
| [SkillPort](https://github.com/gotalab/skillport) | Skill delivery through CLI or MCP across agents | Overlap on portable access |
| [SkillNote](https://github.com/luna-prompts/skillnote) | Registry, synchronization, and agent usage reviews | Direct overlap on feedback |

The comparison does not imply that the existing tools are vendor-locked.
Several explicitly support multiple agents and are open source.
Workshop's design preference is a Git-backed, source-independent record of decisions and experience, with replaceable integrations and project autonomy.

Adjacent research worth acknowledging, not implemented Workshop integrations:

- [SkillHone](https://arxiv.org/abs/2606.08671): persistent decision histories for continued skill improvement.
- [ACES](https://arxiv.org/abs/2608.20614): repository-native paired trials with and without a target skill.

## Slide 3: personal experience and counts

The qualitative statement paraphrases John's report while planning this talk: discoverability and skill usage have improved for him.
It is not a verbatim quotation or a controlled before-and-after result.

Counts come from JSON records under [Workshop memory at dc5adeb](https://github.com/leej3/skills-workshop/tree/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/memory):

- 25 skill records and five project records.
- 110 events: 43 membership, 37 use, 24 consideration, and six contribution.
- Zero evaluation records.

These count remembered records, not all searchable skills, installations, unique people, verified benefits, or controlled experiments.
The same skill can have several use observations.
See [snapshot.json](snapshot.json).

The portable ownership model is documented in the [revision-specific README](https://github.com/leej3/skills-workshop/blob/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/README.md).
Vendor independence is an architectural property and aim, not evidence of identical behavior across all agent hosts.

## Slides 4 and 5: current discovery and feedback

- [Workshop README at the snapshot](https://github.com/leej3/skills-workshop/blob/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/README.md): recall, expanded search, feedback, and insights commands.
- [Registered sources](https://github.com/leej3/skills-workshop/blob/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/registry.toml): K-Dense, NiPreps, CON, Anthropic, OpenAI, and configured installed skill roots.
- [Feedback instructions](https://github.com/leej3/skills-workshop/tree/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/.agents/skills/workshop-feedback): short observations after actual use, optional ratings, attribution, and explicit uncertainty.
  No lifecycle hook ships in this snapshot.
- [CON's REUSE skill](https://github.com/con/skills/tree/master/introduce-reuse-compliance): the concrete demo candidate.
- [Recorded REUSE observation](https://github.com/leej3/skills-workshop/blob/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/memory/events/2026/09/20260909T140926Z-b4c836e1-776d-4b61-b3b9-fdb44d09e68a.json): source for the condensed example, including the Python compatibility caveat.
  The observation is agent-authored and unreviewed.

## Slide 6: collaboration proposal

[con/skills](https://github.com/con/skills) is the existing shared skill source.
The group pilot is a proposal, not an approved CON policy or a deployed service.

The baseline is usage discussions and code review in the existing repository.
[SkillNote](https://github.com/luna-prompts/skillnote) provides another route through a registry, synchronization, and agent reviews.
A group considering it should test hosting and history export against its needs; this presentation does not claim that full export is impossible.

The Workshop option would share reviewed, sanitized observations through Git, potentially in a companion memory repository linked to `con/skills`.
Local use would inform shared records; useful findings would feed skill PRs.
Attribution, identity across forks, duplicate handling, review, and the cost of participation remain practical questions for the pilot.

The suggested few-member, two-task, four-week trial is a recommendation.
It is not a description of an existing group program.