Skip to content

Context and environment

inspect_context collects bounded observations for a selected workspace so you can choose the right target and understand its environment before acting. It is not an agent memory database, a monitoring daemon or an authorization grant.

Configure and select

Use an existing configured local workspace or a prepared, identity-pinned SSH workspace. No additional configuration flag is needed. Choose only the sections relevant to your question:

Section What it reads
workspace One Git status page and target identity.
worktrees Registered worktrees, not proof that they are idle.
activity An explicitly selected task and original operation index.
runtime Controller provenance, configured grants and supported routes.
environment Target observer and paths of explicitly requested executables.
project Bounded root declarations and lock fingerprints, without evaluating them.
resources Supported CPU, memory, filesystem and display-adapter observations.
services Listener counts for explicitly selected TCP ports, without connecting.

Default sections are workspace, activity and runtime. Supply a known task_id for task activity; no selected task does not mean there is no activity.

Example

This is a tool call shape, not a terminal command:

inspect_context(machine="local", workspace_name="project",
  sections=["workspace", "environment", "project", "services"],
  executables=["python3", "git"], ports=[8000])

Or ask ChatGPT:

Inspect the project workspace and resolve python3 and git without running
either program. Check only whether TCP port 8000 has a listener. Do not connect
to it, run project code, install anything or change the machine.

Verify the observation

Read the target identity, per-section source, timestamps, status and complete flag. Follow returned next references when a page is incomplete. An executable path is not its tested version. Listener absence is not port reservation, UDP coverage or service health. Runtime metadata describes the controller, while opt-in environment data describes the selected target observer.

Limits and errors

Sections are not an atomic snapshot. Resource totals are not reserved capacity; GPU metadata is not a utilization or complete compute-device inventory. Older SSH workers may not support environment observation. Preserve that error instead of substituting local facts or silently installing a new worker.

The environment Skill owns detailed collection limits and source semantics.