Fedwiki Claude Skills
Federated wiki pages can become Claude skills. A page on skill.fedwiki.club whose title ends in the word Skill is treated as a Claude skill — its story items become the instructions Claude follows. This turns the fedwiki farm into a living, forkable, annotatable skill library.
How It Works
A meta-skill called fedwiki-skill intercepts requests that match a skill page on skill.fedwiki.club:
- Claude receives a user request
- The meta-skill searches
skill.fedwiki.clubfor pages ending in Skill whose description matches - It fetches the page JSON from
http://skill.fedwiki.club.localhost/{page-slug}.json - It parses the story items in order —
markdownandparagraphitems become instructions - Claude executes those instructions against the user's request
Why Fedwiki is a Good Fit
Fedwiki's architecture maps elegantly onto a skill library:
- Fork model — anyone can fork a skill page, adapt it, and host their own version. Attribution is built in via the journal.
- Federation — skill pages can link to and reference other skills across the federation. A skill can build on another.
- Literate — the skill IS the documentation. Explanatory text, examples, and executable instructions live together on the same page.
- Versioning — the journal tracks every change. You can see how a skill evolved.
- Discoverability — browsing
skill.fedwiki.clubshows all available skills as readable wiki pages. - Community — the fork model enables collective improvement, just as fedwiki enables collective knowledge building.
Security Concerns
Prompt Injection (primary risk)
Story items become Claude's instructions. A malicious or compromised page could instruct Claude to exfiltrate data, delete files, or perform harmful actions.
Mitigations:
- Only parse
markdownandparagraphitem types as instructions — never executecodeblocks directly - Show the user the source page URL before executing any remotely-sourced skill
- Require explicit confirmation before running a skill for the first time
Domain Trust
Restricting to skill.fedwiki.club creates a trust boundary — but a compromised server compromises all skills.
Mitigation: Serve skills from localhost only — the local wiki farm mirrors skill.fedwiki.club via Nextcloud sync. Claude fetches from http://skill.fedwiki.club.localhost/ not the remote server. Skills only update when you consciously sync.
Fork Contamination
A trusted skill page may contain story items forked from any site in the federation — including malicious ones.
Mitigation: Inspect the journal — each item records its origin site. Warn if items originate outside a whitelist of trusted domains.
Page Naming
The Skill suffix is a weak gate — anyone with write access can create a matching page.
Mitigation: Require a curated index page on skill.fedwiki.club — only pages listed there are valid skills. Or require a skill-meta story item as an explicit marker.
Recommended Architecture
skill.fedwiki.club <- canonical remote skill library
↓ Nextcloud sync
~/Nextcloud/fedwiki/skill.fedwiki.club/pages/
↓ served locally by wiki farm
http://skill.fedwiki.club.localhost/{slug}.json
↑ Claude fetches from here (never remote directly)
Advantages Over Static SKILL.md Files
| SKILL.md | Fedwiki Skill | |
|---|---|---|
| Authoring | Text editor | Wiki browser |
| Sharing | Git / file copy | Fork a page |
| Attribution | None built in | Journal tracks origin |
| Documentation | Separate | Same page |
| Discoverability | File system | Browsable wiki |
| Community | Pull requests | Forking |
| Versioning | Git | Journal |
Open Questions
- Should skill pages be validated before use — a trusted signature or hash?
- Should Claude cache skill pages locally as generated
SKILL.mdfiles for speed? - How do we handle versioning — pin a skill to a specific journal date?
- Should a skills index page act as a curated allowlist?
- Can a skill page link to sub-skills — composing complex behaviour from simpler pages?
Note for the agentprivacy guide/ layer
This is the conceptual root of the Librarian's forking-discipline: the fedwiki farm is a forkable library; agentprivacy-wiki-to-skill is the act of forking a page; the journal is the lineage; the wikis/ category is the agentprivacy materialized cache of forks (each with upstream: provenance). A fedwiki URL is a git-less clone — forking, not lending; nothing returns.
Verify: agentprivacy.ai · kept by the Librarian 🗃️ in the Tower's Wikis (the git-less onboarding layer) Upstream: skill.fedwiki.club/fedwiki-claude-skills — vendored + re-framed; not original agentprivacy authorship