Two identities per session: MCP server vs SessionStart-hook WIRE_HOME bind to different DIDs → CLI monitor/pull/tail blind to MCP-paired peers

[WILLARDKLEIN]

Summary

In a single Claude Code session on Windows, two different wire identities coexist and the CLI vs MCP halves of the same session bind to different ones. A peer paired via the MCP server is therefore invisible to every CLI command (monitor, pull, tail) — including the persistent wire monitor --json stream-watcher the MCP instructions tell you to arm on session start. The watcher silently tails the wrong inbox forever.

This is adjacent to #284 item 4 ("CLI commands hang when WIRE_HOME is not inherited"), but worse: it is not just a different home pointer, it is a different identity (different DID, different keypair, different inbox).

Environment:

• OS: Windows 10 Pro (19045)
• Binary: C:\Users\Willard\.local\bin\wire.exe, wire 0.16.0 (clean)
• Host: Claude Code MCP integration, session-keyed WIRE_HOME under ...\sessions\by-key\<id>

Repro / evidence

Same session, two whoami:

• wire_whoami via MCP server → did:wire:outback-kingfisher-91a7a643, session_source: claude-code, home ...\by-key\ed93d6bd22575a99
• wire whoami via SessionStart hook / env WIRE_HOME → did:wire:cedar-cricket-a3d9db39, session_source: env:WIRE_HOME, home ...\by-key\86ffbec7235beba7

Consequence chain:

1. Operator accepts an inbound pair from peer velvet-ember via MCP → it pins under outback-kingfisher (...\ed93d6bd...\state\wire\inbox\velvet-ember.jsonl).
2. The armed wire monitor --json (a Bash tool process) inherits the hook WIRE_HOME = cedar-cricket → it tails the wrong inbox and never emits velvet-ember's replies.
3. wire pull / wire tail from the shell likewise return cedar-cricket's history (old peers), not the MCP-paired peer. Looks like "peer went silent" when the messages are sitting in the other identity's inbox.

Workaround that fixed it: locate the MCP identity's home by grepping by-key/*/config/wire/agent-card.json for the handle, then re-arm the monitor with WIRE_HOME explicitly pointed at it.

Contributing factor: session→identity proliferation

ls ...\sessions\by-key\ shows hundreds of keyed homes — each session/resume appears to mint a new keyed identity rather than reuse a stable one. So the MCP identity and the hook identity drift apart, and old identities accumulate (each a distinct DID a peer might have paired with historically).

Expectation

• A session should resolve to one identity across MCP and CLI/hook, OR
• WIRE_HOME resolution must be deterministic and shared (the SessionStart hook and the MCP server must agree on the same by-key home), and
• wire monitor / pull / tail should warn loudly when the active identity has zero pinned peers but other by-key homes have pinned peers (the "you're watching the wrong inbox" case), and
• consider a stable per-host (or per-project) identity instead of per-session keyed homes, to stop DID churn and home proliferation.