bounded-observe
Installation
SKILL.md
Bounded Observe — Action Boundaries
Action Boundaries are a facet of Boundaries — alongside the data, backend,
and UI boundaries your policy already declares — applied to the external
actions your app or agent takes: ctx.ai spend, ctx.services tool calls,
and agent egress. Give your agent's actions the same proven boundaries your
data already has — watch first, then enforce with one click.
Every action boundary lives in one of three states, and the trust level differs per state. Keep them straight in every answer:
| State | What it is | Trust level |
|---|---|---|
| Watching | Every recognized external action lands in your app's Feed, with per-actor attribution. No behavior change. | Reported. Never imply a watched action is protected. |
| Suggested | Baselines derived from your real events become boundary cards; each card states its evidence (observation window + event count). | A proposal, not protection. |
| Enforced | Promote turns a suggestion into a live boundary; each matched action gets a deterministic verdict. | A native invariant is proven in your app's policy; an escorted action boundary is checked at the platform edge. |