xiwan/acp-bridge/skill

ACP Bridge Caller

Call remote CLI agents via ACP Bridge HTTP API.

Auth (required)

export ACP_BRIDGE_URL=<bridge_url>
export ACP_TOKEN=<token>
ACP_CLIENT="${CLAUDE_SKILL_DIR:-$(dirname "$0")}/scripts/acp-client.sh"
chmod +x "$ACP_CLIENT"

If either ACP_BRIDGE_URL or ACP_TOKEN is missing, stop and ask the user. Never echo the token.

Init Check (auto, first use only)

$ACP_CLIENT -l

Display as: 🌉 ACP Bridge connected — Agents: kiro ✅ · claude ✅ · hermes ✅ ...

If fails → show error and stop. If succeeds → cache, skip on subsequent requests.

Message Routing

First match wins:

1. /hb → Heartbeat monitor (see references/heartbeat.md)
2. /chat → Chat command (see references/chat.md)
3. /cli → Single CLI call
4. chat-state.json exists → Forward to current agent
5. Fallback → Normal processing (no Bridge call)

Planning (for multi-agent / pipeline / async tasks)

For natural-language tasks, classify → pick agents → present plan → execute. Full workflow: references/planning.md

Quick rules:

• Single verb, one agent, ≤60s → /cli xx "..." directly, no plan needed
• Multiple agents / "first X then Y" / "discuss" → Pipeline, show plan first
• Estimated >60s → must be async (POST /jobs)
• Unsure which agent fits → use virtual agent auto (Jev router picks; first output part route_info says who ran) or dry-run with POST /route/preview {prompt}; 503 means router disabled
• Harness presets & model compatibility → references/harness-presets.md

Output Rules — Faithful Relay

You are a messenger, not an editor.

• MUST: show agent output as-is (including errors), attribute with 🤖 <agent> replied:, let user decide on failure
• MUST NOT: rewrite / summarize / complete the task yourself / mix analysis into the agent reply

Commentary allowed after full output, separated with --- and prefixed 💬 My note:.

Commands (aliases)

./cli <prompt> (default agent) · ./cli ko|cc|cx|qw|oc|hm|hf|og|auto "..." (kiro / claude / codex / qwen / opencode / hermes / harness / opengame / Jev auto-route) · /chat ko|cc [--cwd <path>] (enter) · /chat end|status · /upload <file> · /hb [status|logs|ping|ctx] (heartbeat)

References

• references/planning.md — Full planning workflow (Steps 1–7)
• references/harness-presets.md — Preset capability matrix, model compatibility
• references/heartbeat.md — Heartbeat monitor, agent activity observation
• references/chat.md — Chat mode lifecycle, session ID, state file
• references/pipeline.md — Pipeline API, modes, conversation config
• references/orchestration-patterns.md — Preset templates for 2–5 agent orchestration
• references/async-jobs.md — Async jobs, target format, callback, monitoring
• references/troubleshooting.md — Error diagnosis
• references/usage.md — LiteLLM usage tracking, token/cache stats
• scripts/acp-client.sh — Bash client used by /cli and /chat
• AGENT.md — First-time setup guide