You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A growing set of tools now ship their own skills as part of install/setup — writing canonical files to ~/.agents/skills/<name>/ (the shared "agents" convention, tracked via ~/.agents/.skill-lock.json) and placing symlinks in each detected AI agent's target dir (~/.claude/skills/, ~/.continue/skills/, ~/.codex/skills/, etc.).
Different install mechanisms (npx postinstall, npx skills add, Claude Code plugins), same end state: canonical files in ~/.agents/skills/, symlinks in agent target dirs, entries in ~/.agents/.skill-lock.json.
Conflict with skillshare
Skillshare's model: ~/.config/skillshare/skills/ is source of truth, target dirs are synced copies, "never edit targets directly." Bundled skills break this:
They write straight to ~/.agents/skills/ + agent target dirs, bypassing skillshare source
skillshare audit flags them as local/unmanaged
Adding exclude: firecrawl* silences audit but then only the subset of agents the installer detected gets the skills (firecrawl wrote to Claude + Continue only, missed Codex/Gemini/Pi which skillshare does manage)
If the user moves them into skillshare source manually, a future <tool> setup skills (or re-install) silently overwrites the target-dir symlinks, drifting from source
Is there an idiomatic skillshare way to claim target-local skills today that I'm missing? collect looked close but I wasn't sure whether it handles the lock-file pruning piece or cross-agent symlink cleanup.
Would skillshare doctor --suggest-adopt (detecting these and offering to absorb them) be a lighter alternative to a dedicated command?
Context
A growing set of tools now ship their own skills as part of install/setup — writing canonical files to
~/.agents/skills/<name>/(the shared "agents" convention, tracked via~/.agents/.skill-lock.json) and placing symlinks in each detected AI agent's target dir (~/.claude/skills/,~/.continue/skills/,~/.codex/skills/, etc.).Confirmed examples in the wild:
firecrawl/clinpx -y firecrawl-cli@latest initfirecrawl/cli,firecrawl/skills)googleworkspace/cli(gws)brew install googleworkspace-cli, thennpx skills add https://github.com/googleworkspace/cliobra/superpowers-marketplacebacknotprop/plannotatorDifferent install mechanisms (npx postinstall,
npx skills add, Claude Code plugins), same end state: canonical files in~/.agents/skills/, symlinks in agent target dirs, entries in~/.agents/.skill-lock.json.Conflict with skillshare
Skillshare's model:
~/.config/skillshare/skills/is source of truth, target dirs are synced copies, "never edit targets directly." Bundled skills break this:~/.agents/skills/+ agent target dirs, bypassing skillshare sourceskillshare auditflags them as local/unmanagedexclude: firecrawl*silences audit but then only the subset of agents the installer detected gets the skills (firecrawl wrote to Claude + Continue only, missed Codex/Gemini/Pi which skillshare does manage)<tool> setup skills(or re-install) silently overwrites the target-dir symlinks, drifting from sourceProposed:
skillshare adoptWould:
~/.agents/skills/(or target dirs) not already in skillshare source~/.config/skillshare/skills/~/.continue/skills/)~/.agents/.skill-lock.jsonbysource/sourceUrlprefixskillshare syncQuestions
~/.agents/skills/convention specifically?collectlooked close but I wasn't sure whether it handles the lock-file pruning piece or cross-agent symlink cleanup.skillshare doctor --suggest-adopt(detecting these and offering to absorb them) be a lighter alternative to a dedicated command?