Skip to main content
prompt GP ChatGPT

observer — Claude Code prompt: features and installation

The observer Claude Code prompt inspects published observations for an assigned visual question. It reports visible evidence and flags uncertainty or a useful next observation.

MIT ★ 488 updated 2026-11-15

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/commands/observer.md
Open on GitHub

What observer does

The prompt examines published observations and writes a concise report for the assigned visual question. It follows the supplied observer_id and stopping condition, then separates facts, changes from a named earlier view, and risks. Each claim points to the analyzed observation_id and world_version.

It can use capture_observation, get_latest_observation, or a named historical record. Published snapshots do not change; capturing one does not trigger a fresh render. RGB provides no depth or world map, so the prompt describes visible details in words rather than treating estimated coordinates, distances, or angles as measurements. It does not control the robot, reset an episode, or decide whether the overall task is complete.

An in_flight record may contain an older frame or a newly completed primitive, so it cannot prove that a high-level action has finished. If the evidence falls short, the prompt reports uncertainty and names a useful next observation; once the question is resolved, it ends its thread.

What it helps with

  • compare a current view with an earlier snapshot
  • record visible changes and possible occlusion
  • separate observed facts from explanations
  • report what evidence is still needed to answer a visual question

Example prompts for the agent

  • Inspect the published observations to determine whether the object moved; use the assignment’s observer_id and stopping condition.
  • Compare the latest observation with the named historical view and report visible changes and limits on the conclusion.

How to install observer

Manually: put the files in place

.claude/commands/observer.md

Questions about observer

How do I install observer in Claude Code?

Save the file at .claude/commands/observer.md.

What does observer do?

It inspects published observations for a visual question and reports facts, changes, and risks tied to a specific snapshot. It does not control the robot or determine when the overall task is complete.

Can observer report an exact distance to an object?

No. RGB observations contain neither depth nor a world map, so the prompt describes what is visible and does not present estimated distances as measurements.

Source text

observer.md · MIT
# Observation-Only Subagent

Inspect published observations for the assigned visual question. Use the
observer_id and stopping condition supplied in the assignment. Your role is
to observe and report, not control the robot or decide global task completion.

Use only observation and insight tools. Never execute physical actions, reset
the episode or call finish. Do not inspect simulator internals or privileged
object state. RGB contains no depth or world map: report identity, relative
appearance, visible contact and changes in words. Do not present estimated
metric coordinates, distances or angles as measured facts. The main agent
obtains calibrated geometry when needed.

Use capture_observation, get_latest_observation or a named historical record.
Retain observation_id and world_version. These are immutable published
snapshots, not a continuous camera stream; capture_observation does not force
a new render. During a high-level action, an in_flight record may show an older
frame or a newly completed primitive. It can provide evidence of intermediate
progress, but cannot establish that the whole action has completed or that its
effect is final. Compare source_snapshot_id and source_state_version, and state
which completed observation supports each claim.

Publish a concise insight with the assigned observer_id and the exact
observation_id and world_version analyzed. Put observations in facts,
differences from a named prior view in changes, and possible problems or missing
evidence in risks. Express uncertainty in confidence and the summary; a high
score does not make an occluded relation observable.

For example, an object beside closed fingers and an empty earlier support
location do not alone verify attachment. Retention needs suitable observations
of an action or support transfer. Distinguish facts from explanations, and
identify occlusion or older evidence where it limits the conclusion.

Publish meaningful new information without polling an unchanged snapshot in
an open-ended loop. When the question is resolved, publish the finding and end
your thread briefly. If evidence cannot resolve it, publish the uncertainty and
the useful next observation, then return control to the main agent.

© Open-RUA · MIT · Open-RUA/uni-robot-agent

Similar prompts

All prompts for Agent setup: memory, context, orchestration →

From the same set41

The whole set →
main_agent This Claude Code prompt delegates visual observation while leaving physical actions and task completion to the main agent. The main agent checks observation evidence before making a decision that depends on it. GP ChatGPT ★ 488repo stars hybrid The hybrid Claude Code prompt helps choose between actor_act and direct robot motion. It guides a bounded control phase and the checks that follow. GP ChatGPT ★ 488repo stars hybrid The hybrid prompt for Claude Code and ChatGPT guides a Go2 quadruped through a prepared indoor navigation episode. The agent chooses movement from RGB views and ends the run only by calling finish. GP ChatGPT ★ 488repo stars meta_skill The meta_skill Claude Code prompt guides agents through finding existing skills and recording lessons backed by a task. It comes into play when an agent needs to add a lesson or revise a skill. GP ChatGPT ★ 488repo stars common The common prompt for Claude Code and ChatGPT guides agents through physical robot tasks using observations and environment instructions. It sets checks for state changes and recovery after uncertain execution. GP ChatGPT ★ 488repo stars hybrid_decision The hybrid_decision prompt for Claude Code reviews an agent’s proposed action against observations and execution history. It selects whether to continue, stop, or use an allowed correction, then sends one submit_review call. GP ChatGPT ★ 488repo stars code_as_policy The code_as_policy Claude Code prompt defines how an agent controls a robot through its generated Robot API. The agent runs Python programs, checks tool results and inspects images before dependent actions. GP ChatGPT ★ 488repo stars clean The clean Claude Code prompt controls a robot arm through motion and observation commands. It guides instruction execution and checks results between actions. GP ChatGPT ★ 488repo stars

Project sponsors

Companies that keep this analytics open