skillshare
Sync and manage AI skills and agents across multiple CLI tools and environments from a single source of truth.
Install
mkdir -p .claude/skills/skillshare && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/2559" && unzip -o skill.zip -d .claude/skills/skillshare && rm skill.zipInstalls to .claude/skills/skillshare
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Syncs skills across AI CLI tools (Claude, Cursor, Windsurf, etc.) from a single source of truth.
Use when: "sync skills", "install skill", "search skills", "list skills", "show skill status",
"backup skills", "restore skills", "update skills", "new skill", "collect skills",
"push/pull skills", "add/remove target", "find a skill for X", "is there a skill that can...",
"how do I do X with skills", "skillshare init", "skillshare upgrade", "skill not syncing",
"diagnose skillshare", "doctor", or any skill/target management across AI tools.Key capabilities
- →Sync skills across global and project directories
- →Install and track skills from Git repositories
- →Manage AI tool targets and filters
- →Audit skills for security and spec compliance
- →Backup and restore skill configurations
How it works
It manages symlinks and file copies between a central source of truth and various AI CLI tool directories.
Inputs & outputs
When to use skillshare
- →Sync skills across different AI tools
- →Backup and restore skill configurations
- →Install new skills from repositories
- →Troubleshoot broken symlinks or sync issues
About this skill
Skillshare CLI
Use Skillshare to maintain a source and distribute resources to configured targets.
Read only the reference relevant to the requested operation; the routing table is below.
Check skillshare <command> --help if the installed version differs from these examples.
Scope and resource selection
- Global config:
~/.config/skillshare/config.yaml. Project config:.skillshare/config.yamlorskillshare/config.yaml; hidden wins when both exist. Use-gor-pexplicitly when location matters instead of relying on auto-detection. - Skills are directories containing
SKILL.md; agents are single Markdown files. Source paths can be customized. Use the configured sources instead of assuming defaults. - Tracked
skills/<known-agent>/<skill-path>/SKILL.mdcopies form one logical skill per repository/prefix/suffix. Merge/copy select the exact agent variant (including aliases and accountagent:identity). Recognition requires the same suffix under two distinct canonical agents. Compatible targets sharing a folder select the union of their agent identities; a universal-only target does not infer an agent. Status counts logical skills and physical variants; lists and sync operations retain physical names. Explicit targets/overrides and physical filters still apply. Whole-directory symlink mode exposes all variants. See sync.md. - Use native agents targets for agents. Use extras for arbitrary file resources or when
flattening/content transformation is required;
extension:works only on extras targets. - Shared memory notes live in the
memoryextra. Useextras memoryto manage Markdown notes and reading guidance; nested paths such aswiki/architecture.mdare supported. Read extras.md for scope, version checks, and agent setup. For a session retrospective or Dreaming across saved notes, read memory-review.md. The host agent performs the review; Skillshare never launches or schedules an agent. - MCP uses its own receiving targets and source definitions. Read mcp.md
before editing MCP settings or importing native configurations. Its Agent notes cover
what is specific to Claude Code, Codex, OpenCode and Pi: scopes and files, names and
credentials they refuse, Pi's built-in MCP and
piOptions, OMP's native MCP and preserved enable/disable lists, the portabletoolspolicy, upgrading Pi settings from 0.22, turning a global server off in one project,mcp.projectsfor many folders from one config, andmcp servefor Agents that cannot reach the synced skill folders. - Plugins keep their native components together. Read plugins.md for installation, import, sync selection, updates, and native compatibility limits across Claude, Codex, Cursor, Antigravity, Pi, OMP, and OpenCode, including Pi packages from pi.dev and OMP marketplace plugins and standalone extension selection.
- Hooks preserve each Agent's native event map or extension/plugin code. Read hooks.md before import, synchronization or recovery.
Execution rules
- Choose noninteractive inputs. Supply names and selection flags supported by the
specific command. Use
--no-tuifor plain output or--jsonwhere supported. Do not add--forcesimply to avoid a prompt: it can overwrite conflicts or override audits. - Keep previews separate from writes. Use
--dry-runwhere supported. After changing skills or agents, sync the intended resource and scope. MCP and hooks mutations save source only unless--syncis supplied;sync --allalso includes extras, MCP and hooks, but excludes plugins. - Inspect audit blocks. Review findings before choosing an override.
install --jsonselects all when no skill/agent filter is given, preserves existing items unless--forceis supplied, and retains the audit gate; it is not merely an output format. Read install.md and audit.md before using it for automation. - Use the matching recovery mechanism. Uninstalled skills go to trash; native MCP entries use MCP backups. Do not substitute filesystem deletion for CLI uninstall.
- Keep credentials out of MCP definitions. Use
{fromEnv: VARIABLE}references. Managing settings does not require resolving secrets, copying OAuth credentials, starting servers, or probing their tools. - Respect operation scope. Local configuration work does not imply permission to commit, push, or publish. Run those commands when included in the user's request.
Common workflows
Inspect and diagnose
skillshare status --json
skillshare list --json
skillshare diff --json
skillshare doctor --json
skillshare mcp --json
doctor exits 1 when it finds errors. Audit uses --format json; JSON flags and
TUI availability vary by command. See the relevant reference rather than assuming parity.
For automatic release checks and notices, use skillshare updates off / on.
skillshare updates check explicitly checks the CLI version even when automatic
checks are off; see status.md for the shared preference
and environment override. Installed resource-source checks are separate.
Track an upstream repository or install selected skills
Choose the resource and scope before installing. Use -g for user-wide resources
or -p for this project; keep that scope on preview, apply, update, and sync.
| Intent | Command |
|---|---|
| Keep an upstream skill collection as one Git checkout | skillshare install owner/repo --track --name team -g |
| Install only named skills with their source metadata | skillshare install owner/repo -s pdf,commit -g |
| Install all discovered skills individually | skillshare install owner/repo --all -g |
| Manage a complete native plugin, including hooks/MCP | Use plugin discover and plugin add; see below |
For the scoped preview → apply → update → sync workflow, read install.md. It covers following the latest upstream branch, rejected flag combinations, and target filters.
For project setup, use init -p, then install -p and sync -p. Rehydrate configured
remote dependencies with install without a source. For full flag details and agents,
read install.md and sync.md.
To use a local skills checkout in place, run skillshare link <path> --enable, then
sync; skillshare unlink <name> moves only the link to trash and, in global mode,
forgets its tracked-repository record and audit acceptances. A metadata warning means the link
is gone but its record may remain; resolve the metadata error before installing again. Its target stays untouched. Read
TROUBLESHOOTING.md for how such links are followed.
Disable or remove
skillshare disable 'draft-*' --dry-run
skillshare disable 'draft-*'
skillshare enable 'draft-*'
skillshare uninstall my-skill
skillshare sync
skillshare trash restore my-skill
skillshare sync
Run the requested operation, not this entire example sequence. Quote glob patterns so Skillshare receives them unchanged. Disabling uses ignore rules; uninstall of installed content uses trash. A global tracked declaration without a checkout is removed as metadata only, with no trash undo; see install.md for canonical names, occupied paths, and the limit after reconcile.
Move skills into folders
skillshare move my-skill frontend --dry-run
skillshare move my-skill other-skill frontend
skillshare sync
Use move instead of mv for installed skills: it keeps their install records, so
nothing is downloaded again. It does not sync; run sync afterwards. Read
install.md for refusals and project mode.
Project folders from the global config
skillshare project list --json
skillshare project localize ~/work/existing --dry-run
skillshare project add ~/work/app --target claude,codex --skills-mode copy --dry-run
skillshare project add ~/work/app --target claude,codex --skills-mode copy --sync
skillshare project add ~/work/suite --scan . --scan .worktrees --target codex --dry-run
skillshare project edit app --skills-include 'frontend-*'
skillshare project sync app --json
skillshare project remove app --dry-run --json
project always uses the global config, including inside a repo; it does not accept
-p. New projects are saved in config.yaml, or on this machine only with
--machine-local or projects_storage: machine-local (--shared overrides the setting);
localize and share move one between the two. Removing a local copy of a config.yaml
project keeps config.yaml's version unless --shared is given. Edit filters replace lists. Removal detaches skills and keeps agent files;
JSON removal requires --yes unless previewing. --scan DIR also adds each Git checkout
below DIR with the same settings and only adds; rerun it after a new clone or worktree.
Read targets.md
for saved-then-sync failures, retries, and partial removal results.
MCP settings
skillshare mcp import --from claude --json
skillshare mcp add docs --url https://example.com/mcp --target claude --no-tui
skillshare sync mcp --dry-run --json
skillshare sync mcp
--target is singular and repeatable. A preview does not apply changes. For revisions,
editing, import conflicts, credentials, backups, and human-operated TUI commands, read
mcp.md.
Complete plugins
Use plugin discover and plugin add for a complete native package; plugin import
adopts an existing native installation. Run only the requested operation. Read
plugins.md before applying changes; it covers th
Content truncated.
When not to use it
- →When manually deleting symlinks
- →When bypassing security audits without force
Prerequisites
Limitations
- →Node extensions require specific environment handling
- →Extensions only supported on extras targets
How it compares
It provides a unified management layer for multiple AI tools rather than configuring each tool independently.
Compared to similar skills
skillshare side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| skillshare (this skill) | 1 | 5mo | Review | Intermediate |
| subtask | 3 | 9mo | Review | Advanced |
| monitor-tweet | 0 | 7mo | Review | Advanced |
| agentic-jujutsu | 3 | 8mo | Review | Advanced |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
subtask
zippoxer
Parallel task orchestration CLI that dispatches work to AI workers (via Claude Code) in isolated git workspaces. Use when the user wants to draft, create, run, or manage tasks, delegate tasks to workers/subagents, or mentions subtask or Subtask.
monitor-tweet
ArthTech-AI
Run the tweet-to-toolkit pipeline on a tweet. Flow: (1) triage+feasibility, (2) ALWAYS show findings and BOTH repo options to user, (3) user approves target + direction, (4) build auto-starts, (5) review auto-starts, (6) if APPROVED, gh pr create auto-runs. The ONLY human gate is step 2 — everything
agentic-jujutsu
ruvnet
Quantum-resistant, self-learning version control for AI agents with ReasoningBank intelligence and multi-agent coordination
moai
modu-ai
MoAI super agent - unified orchestrator for autonomous development. Routes natural language or explicit subcommands (plan, run, sync, fix, loop, project, feedback) to specialized agents. Use for any development task from planning to deployment.
build-with-agent-team
coleam00
Build a project using Claude Code Agent Teams with tmux split panes. Takes a plan document path and optional team size. Use when you want multiple agents collaborating on a build.
swarm-coordination
joelhooks
Multi-agent coordination patterns for OpenCode swarm workflows. Use when work benefits from parallelization or coordination.