Vendor Support¶
One shared memory, every major AI tool. "AI enable this repo" detects whatever AI footprint
already exists, folds it into the shared layer, and routes that vendor's agent to the one hub
(AGENTS.md) — originals preserved, never deleted.
How neutrality works¶
agent-memory bets on open standards:
AGENTS.mdfor steering — the hub every agent reads first.- The open Agent Skills standard for capabilities — one
SKILL.mdshape.
Tools that adopt them plug in with little or no glue, which is why new vendors (Kiro, Copilot, Antigravity) have shipped as point releases.
Detected & migrated¶
A vendor's steering folds into memory/; checked-in chat history (e.g.
.aider.chat.history.md, .continue/sessions/) becomes dated session logs; a vendor's skills
are promoted into the neutral agent-skills/ layer. Anything in your home folder stays in
your home folder.
Supported footprints include: Claude Code, Cursor, GitHub Copilot, Gemini CLI, Google Antigravity, Amazon Kiro, GPT / Codex, Cline, Roo Code, Aider, Continue.dev, Codeium / Windsurf, Zed AI.
Native skill adapters¶
For runtimes with a native skill/command system, sync skill adapters regenerates thin
pointers across six targets (gitignored, never copies):
| Vendor | Adapter path |
|---|---|
| Claude Code | .claude/skills/<name>/SKILL.md |
| Gemini CLI | .gemini/commands/<name>.toml |
| Cursor | .cursor/rules/<name>.mdc |
| Amazon Kiro | .kiro/steering/ (skill pointer) |
| GitHub Copilot CLI | .github/skills/<name>/SKILL.md |
| Google Antigravity | .agents/skills/<name>/SKILL.md |
Antigravity vs. Gemini CLI
Antigravity (the Gemini CLI successor) reads workspace skills from .agents/skills/,
not the old .gemini/commands/*.toml. The Gemini TOML adapter stays for now so Gemini
CLI keeps working through the transition.
Copilot hot-reload
GitHub Copilot CLI loads .github/skills/ at init, so a freshly-synced skill needs a
/restart (or skills rescan) to appear. Claude / Cursor / Kiro don't.
Vendor-specific notes¶
- Enterprise IDEs (Kiro). Per-machine vendor dirs (
.kiro/,.claude/, …) are gitignored, so a fresh clone won't have them — runsync skill adaptersafter the agent loads the protocol. Human-gated commit hooks (like Kiro's) align with the deliberate-commit model. - GitHub Copilot. Its
copilot-instructions.mdfront-loads the explicitmemory/read list (Copilot's Ask/Plan modes don't reliably chase a pointer chain) and leads with a first-run init block.