Comparison with existing tools
August 21, 2026 · View on GitHub
中文版：与现有工具对比
Measured 2026-08 via the GitHub API (stars) and repository activity (last push). "Active" = pushed within roughly the last 3 months at measurement time; the table is a point-in-time measurement, not a trend claim. Competitor capabilities are stated only where their READMEs or the cited community threads make them explicit; claims we cannot verify are listed as not-verified.
Adjacent categories (proof of demand, not direct competitors)
| Category | Example | Measurement |
|---|---|---|
| Skill content | anthropics/skills | ~170k★, active |
| Multi-agent config/API switcher | farion1231/cc-switch | ~128k★, active |
| GUI skill manager | xingkongliang/skills-manager | ~3.8k★, active |
Direct headless niche: cross-agent config sync
| Repo | Stars (2026-08) | Status | Notable limitation (from their own materials / community reports) |
|---|---|---|---|
| gotalab/skillport | 408 | slowed | — |
| amtiYo/agents | 85 | active | — |
| athola/skrills | 69 | active | — |
| nicepkg/vsync | 55 | no push for ~7 months | setup manual; npm ecosystem full of dead duplicates |
| 2ue/ccman | 52 | — | — |
| xiaolai/cc-suite | 41 | — | — |
| slash9494/ai-config-sync-manager | 28 | — | two-tool scope |
| miniLV/Plexus | 27 | stalled | local one-shot, no cross-machine |
| Leoyang183/sync-agents-settings | 8 | stalled | — |
| berlinguyinca/ai-sync | 37 | — | Claude-only |
| Vek-Sync | 3 | — | MCP-only |
| cortesi/agentsmd | 9 | — | rules-only |
Pattern: the headless "sync agent config" niche is a graveyard of near-identical, mostly single-agent or rules-only tools under ~500 stars. That is the discovery/trust problem any new entrant must solve — and it is why skillferry leads with the portability contract and the security boundary rather than "supports N agents".
What we can write vs. what we cannot write
We can write (evidence anchors):
- Codex and DSH both natively discover
~/.agents/skills/— exercised in the author's macOS and Windows acceptance records (macOS, Windows). - The workspace format is target-neutral; each agent is only a render target (PORTABILITY_CONTRACT).
- Secret references are schema-enforced; the shareable export never expands them; this is machine-verified in CI, not documented intent.
- Every grade is derived from the capability table
(AGENT_MATRIX), including honest
manual/degradedlabels.
We cannot write (not verified — do not claim):
- "Claude Code natively scans
~/.agents/skills/." - Any performance or reliability comparison of competitor tools we have not run (no benchmark exists in this repo).
- "Model skills always trigger" claims: community reports of skills not triggering are user perception, not our measurement.
- Star trends (the numbers above are a single 2026-08 snapshot).
Explicitly not doing (v1)
GUI (the GUI segment has volume but a different shape); provider/API switching; session/memory/history sync; dotfiles wholesale sync; lossless arbitrary plugin conversion; "support N agents" as a headline; SSH targets; team layer (v1.x).