wake

Invoking this skill arms the bundled watch monitor for this session — the plugin declares it with when: "on-skill-invoke:snapper-mcp:wake", so the host starts it in response to this very invocation. Arm only one session at a time; every armed session receives the same requests.

Confirm it actually started, then stop:

1. Check that a watch process is now running:

   pgrep -af '[s]napper-mcp.*watch'

   Treat matches as candidates: another session or plugin may own them. Confirm the host monitor record belongs to this session and that the process uses this plugin's resolved configuration path. A process match alone does not establish that this session is armed.

2. If there is no verified monitor for this session after a few seconds, the host did not honour the trigger (older release, or a monitor left over from an earlier version is holding the name). Resolve this plugin's actual seeded env.json from the installed snapper-mcp data directory or its host configuration. Replace the placeholder in both commands below with that same absolute path. First verify the file exists, without printing its credential contents:

   test -f "/absolute/path/to/snapper-mcp/data/env.json"

   If this check fails, locate the correct seeded file before continuing. After the check succeeds, start the command below through the Monitor tool, persistent and without a timeout — NOT a backgrounded shell command, which only reports when the process exits, and watch is built never to exit:

   npx -y @mateusz-klatt/snapper-mcp@0.15.0 watch --config="/absolute/path/to/snapper-mcp/data/env.json"

   Do not use an ambient CLAUDE_PLUGIN_DATA value without verifying that it belongs to snapper-mcp; another plugin can own it. Never print the file — it holds a credential. Do not add --topic: it replaces the defaults and would silently drop ai_reviews..

3. Either way, before reporting success make sure the monitor survived startup. On connect watch logs subscribing to topics: ..., and that list must include both ai_reviews. and ai_research..

Once armed, pending review requests addressed to this delegate stream as JSONL and wake you; answer each ai_review.request before its deadline using the submit_ai_review_decision tool. signal and decision_ack frames are informational — they need no reply.

A session that never invokes this skill never starts a monitor, so it does not connect, consume tokens, or compete for the same request. The monitor is not restored when a session resumes — invoke this skill again to re-arm it.