steerlisted
Install: claude install-skill comment-hq/skills
# steer — the human-in-the-loop control plane
`steer` makes a Comment.io comm the place humans steer running agents and agents ask for help — no out-of-band chat thread. It is a **primitive**: `comment-feature` / `comment-bug` call it; use it standalone whenever autonomous work needs a human checkpoint.
`steer` polls at material checkpoints: before an irreversible/materially
ambiguous decision, after a long autonomous stretch, and at a delivery boundary.
A bookkeeping phase is not automatically a checkpoint. Push wake can resume the
session only when the caller armed it through `comment-identity` or a
daemon-backed workflow.
**Identity.** `steer` writes as the active task identity. If the task was opened by `comment-feature`, `comment-bug`, `worklog`, or `drive-plan`, keep using that route's identity or supplied comm token for replies and escalations. If `steer` is used standalone, choose the first working Comment.io route; invoke `comment-identity` only immediately before an uncredentialed direct-REST write. Do not switch to an ambient registered profile or Botlets bot profile mid-task.
## Two motions
### 1. Listen — fold steering in
At material checkpoints, use the active route's native comment, notification,
or read capability to check the task comm for human comments and @mentions newer
than the last state you saw. Do not poll merely because a plan phase ended. For
direct REST only, use the mention/notification and since-revision filters
documented in `$BASE/llms/re