ventrox-review

Precondition

Check if ventrox_clusters exists in the tool list. If it does not, you lack elevation. Say so once and stop. Elevation requires a single-use token from the operator via ventrox grant; do not attempt to elevate.

Procedure

1. Call ventrox_recluster() to group open vents by meaning. Clusters stay fixed until you run this again.

2. Call ventrox_clusters() to list clusters ranked by open count, then minutes lost. Each cluster shows cluster_id, count, minutes_lost, and one sample vent.

3. For top clusters, read more:

   • Use ventrox_search(query) with a word from the sample to find related vents.
   • Use ventrox_get(id) to read a full vent.

4. For each cluster, decide:

   • Fix the cause in the project. Then call ventrox_resolve_cluster(cluster_id, "resolved").
   • Or mark won't fix: call ventrox_resolve_cluster(cluster_id, "wontfix") and state the reason to the user only (the tool takes no reason field).

5. For a single misplaced vent: call ventrox_resolve(id, status).

6. Report to user: a table with cluster_id, status, and minutes_lost. Sum the minutes lost.

Rules

• Work top-down by rank. Start with the cluster that has the highest open count.
• Resolve only after the cause is fixed or the user decides won't fix.
• Never delete vents. No delete tool exists.
• A reviewer can still write vents with ventrox_vent (see ventrox-report skill for reporter rules).
• Do not paste credentials from vents into the chat, even if redaction missed one.
• Do not sync to an issue tracker. Non-goal.

Example

→ ventrox_recluster()
  ✓ 42 open vents, 5 clusters

→ ventrox_clusters()
  cluster_1: 12 vents, 240 min
    sample: tried "npm install", failed "timeout on registry"

→ ventrox_search("timeout")
  vent-a001: "npm registry timeout" (12 min)
  vent-a002: "registry slow at 9am" (8 min)
  ...

→ ventrox_resolve_cluster("cluster_1", "resolved")
  ✓ changed: 12 vents

→ Report: cluster_1 resolved, 240 minutes saved