## Summary - ACP runtimes can expose Fusion custom tools (`fn_*`) to external agents such as Hermes ACP and Prime. - When the engine passes `customTools`, `AcpRuntimeAdapter` starts a per-session loopback tool bridge and registers it as a stdio MCP server in `session/new.mcpServers`. - The bridge uses a per-session bearer token, binds only to loopback, preserves the MCP request ID as `toolCallId`, propagates tool `isError` results, and exposes only runnable non-built-in tools. - The MCP shim supports `initialize`, `ping`, `tools/list`, and `tools/call`, reporting transport, authentication, malformed-request, and unknown-tool failures correctly. - Bridge startup failures degrade gracefully: the ACP session remains usable without custom tools and exposes a fixed `fusionToolBridgeError` reason code for engine auditing. - Startup, request, and teardown paths clean temporary schemas and listeners; disposal aborts cooperative tools, bounds non-cooperative drains, prevents post-dispose execution, and preserves the existing synchronous session contract through an awaitable `disposePromise`. - The packaged CLI stages `mcp-schema-server.cjs` beside the ACP runtime distribution. ## Verification Validated at source commit `2fa6fd2e26923ac6241d1b9379b3819f1832c079`: - `pnpm --filter @fusion-plugin-examples/acp-runtime test` — 21 files / 251 tests passed - `pnpm --filter @fusion-plugin-examples/acp-runtime typecheck` — passed - Plugin build with `.cjs` asset copy — passed - `pnpm check:changesets` — passed - `git diff --check` — passed - MCP shim smoke test — passed - Greptile Review — passed - CodeRabbit — passed - Devin review — 0 bugs, 6 analyses, `errored_tasks=[]`, quality `4/10` ## Scope ACP runtime plugin only. The Hermes CLI runtime remains untouched; Hermes ACP is enabled by selecting the generic ACP runtime with `acpBinaryPath: hermes` and `acpArgs: ["acp"]`. --------- Co-authored-by: gsxdsm <gsxdsm@users.noreply.github.com>
@fusion-plugin-examples/acp-runtime
A Fusion runtime plugin that drives any external Agent Client Protocol (ACP) agent over JSON-RPC/stdio. One integration unlocks every ACP-compatible agent (Gemini CLI, the Claude Code ACP adapter, and any future agent that speaks the protocol) through the standard protocol instead of a bespoke per-CLI integration.
Selected via runtimeId: "acp". Installed on demand (experimental) — see the
Fusion plugin catalog (fn plugin install fusion-plugin-acp-runtime).
Security posture
The ACP agent is an untrusted subprocess that calls back into Fusion for permissions and filesystem access. This plugin enforces a defense-in-depth floor:
- Per-category permission gating. Each
session/request_permissionis classified by tool kind into a Fusion action category and checked against the live permission policy — never a preset shortcut.allow_onceonly (never a persisted blanket grant). Unmappable kinds and missing policy default-deny. - Unrestricted-risk acknowledgement (
acpAllowUnrestricted). Because the shipped default policy isunrestricted(allow-all), a blanketallowon a sensitive category is escalated to approval unless the user explicitly setsacpAllowUnrestricted: true. Prefer running the ACP runtime under anapproval-requiredpolicy. - Filesystem jail.
fs/read_text_file/fs/write_text_fileare opt-in (acpFsRead/acpFsWrite, writes default OFF), confined to the sessioncwdby a real symlink-resolving jail (realpath +O_NOFOLLOW), with a deny-list for secrets (.env,*.pem, …) and git internals (.git/**). Writes are gated through thefile_write_deletepermission category. - Untrusted-input bounds. Streamed output is sanitized (ANSI/control strip) and bounded (per-turn + per-chunk caps; bounded tool-call correlation map).
- Subprocess isolation. The agent env is built from an allow-list
(
acpEnvAllowList) — inheritedprocess.envis not forwarded.
Not sandboxed in v1: the agent's own process/network syscalls run with Fusion's user privileges (OS-level sandboxing is recommended future work).
Settings
| Key | Default | Meaning |
|---|---|---|
acpBinaryPath |
acp-agent |
Agent binary to spawn |
acpArgs |
[] |
Args that launch the agent in ACP/stdio mode (e.g. ["--acp"]) |
acpModel |
— | Optional model identifier reported via describeModel |
acpFsRead |
false |
Advertise/register fs/read_text_file |
acpFsWrite |
false |
Advertise/register fs/write_text_file (gated) |
acpEnvAllowList |
[] |
Env var names forwarded to the agent subprocess |
acpAllowUnrestricted |
false |
Acknowledge the untrusted-agent risk under an allow-all policy |
Upstream / third-party integration evidence
Per AGENTS.md (External-integration evidence):
- Protocol homepage / docs: https://agentclientprotocol.com
- Upstream protocol repo: https://github.com/agentclientprotocol/agent-client-protocol
- TypeScript SDK repo: https://github.com/agentclientprotocol/typescript-sdk
- Dependency (npm):
@agentclientprotocol/sdk— https://www.npmjs.com/package/@agentclientprotocol/sdk - Pinned release:
0.24.0(Apache-2.0) - Tarball: https://registry.npmjs.org/@agentclientprotocol/sdk/-/sdk-0.24.0.tgz
- Integrity (sha512):
sha512-vvu9appvGvfYstBj19C6NCepV6SvUhY5VRv60KUZ4XzhTah/olOYul5Zo4C+x2enyshMSvgB2mm/OEmrsHaSmA== - Agent binaries driven: user-supplied ACP agents (e.g.
gemini --acp) and the bundled Claude bridge below. User-configured agents remainupstream-pending-verificationper agent.
Bundled Claude ACP bridge evidence
- Canonical upstream repo URL: https://github.com/moabualruz/claude-code-cli-acp
- Docs / homepage URL: https://github.com/moabualruz/claude-code-cli-acp#readme
- Release / download URL: npm package
claude-code-cli-acp(version0.1.1) — https://www.npmjs.com/package/claude-code-cli-acp - Binary / CLI name:
claude-code-cli-acp - Checksum:
sha512-qpfRGOXkOs9mqI7oumsGistWisyXcCC0r7ng7wdLvGMIORdzHjmUUa+94Jftgr/NYAVnAUe6N7kimD8PaO3D5g==(frompnpm-lock.yamlforclaude-code-cli-acp@0.1.1) - Pinned-commit spot-review: tag
v0.1.1points to commitc93f4f4ca449f451d9f3b7db536caf4060883da9(annotated tagca33404fc1128d6a88a55b248f042f70b4bc9f9a, unsigned). License: Apache-2.0. Behavior reviewed for this integration: runsclaudethrough a PTY, reads transcript JSONL, exposes an ACP server over stdio, and requires@anthropic-ai/claude-codeinstalled + authenticated.
See docs/acp-contract.md for the launch/readiness contract and failure taxonomy.