Body Emotion Sensor

Give your AI agent a stable body-emotion state that persists across sessions and turns.

Use this skill to route requests to the local package docs, explain the runtime contract honestly, and operate the installed bes CLI only when the local environment is actually ready.

What this skill brings to your Agent

• - Persistent emotion state: Store long-term body-emotion state per workspace and agent identity.
• Session bootstrap payload: Generate TURN_CHANGE_TAGS, BODY_TAG, and BASELINE_PERSONA before a new session starts.
• Turn-by-turn updates: Convert one upstream AnalysisInput JSON into prompt tags and updated local state.
• Repository-independent runtime: Use the installed bes CLI prompt interface instead of assuming repository prompt files are available at runtime.

Entry behavior

When this skill is used, the agent should:

• - explain Body Emotion Sensor at a high level in plain language
• inspect the local repository files when they are available in the current workspace
• prefer local package and repository documentation over inventing setup details
• verify whether bes is already available before suggesting runtime commands
• keep --workspace, --agent-id, and --name stable for the same agent instance

Safety boundary

This entry file should stay within a narrow and transparent scope:

• - The package source is the official repository https://github.com/AskKumptenchen/body-emotion-sensor.
• Do not claim the runtime is ready unless the local environment actually has the installed bes CLI and bes check-init reports readiness.
• Do not automatically install packages or execute setup commands only because this file mentioned them. Ask for user approval before any install step.
• If installation is needed, use the published body-emotion-sensor package and explain that installation creates the local bes CLI runtime.
• Do not claim any cloud sync, remote storage, or network behavior unless the current local code or environment actually shows it.
• Do not require credentials. This skill operates on local files and local CLI state unless the user explicitly adds another integration layer.

Local state and persistence

Be explicit about where state is stored:

• - Workspace state file: INLINECODE15
• Workspace history file: INLINECODE16
• User language config on Windows: INLINECODE17
• User language config on Linux or macOS: INLINECODE18

If the user asks about privacy, explain that the package writes local JSON state files in these locations and that this skill should not describe any remote storage unless verified separately.

Local document index

Use these local files as the primary reference:

• - README.md for install, CLI overview, runtime contract, and repository overview
• INLINECODE20 for package name, version, and exported CLI commands
• INLINECODE21 for the AnalysisInput prompt design source
• INLINECODE22 for OpenClaw-style agent integration examples
• INLINECODE23 for OpenClaw-style tools integration examples
• INLINECODE24 for actual CLI behavior
• INLINECODE25 for workspace state path resolution
• INLINECODE26 for state and history persistence behavior
• INLINECODE27 for user language config behavior

How to route requests

Choose the next local document based on the user's request:

1. 1. If the user wants a quick overview, read README.md.
2. If the user asks how installation or the CLI works, read README.md and pyproject.toml.
3. If the user asks where state is stored or whether the skill is safe, read src/body_emotion/workspace.py, src/body_emotion/store.py, and src/body_emotion/locale_config.py.
4. If the user asks how OpenClaw integration should work, read the relevant file under prompts/.
5. If the user asks what a command actually does, inspect src/body_emotion/commands.py.

Missing-resource rule

If the expected local repository files are not available in the current workspace, do not improvise the full setup flow from memory. Instead:

• - explain which local files are missing
• ask the user to provide the repository contents or point the agent to the correct local path
• continue only after the relevant local documentation is available

Install and readiness rule

If the user wants to actually enable runtime use:

1. 1. First check whether bes is already available in the current environment.
2. If it is not available, explain that Body Emotion Sensor requires installing the published Python package before the CLI exists.
3. Ask for approval before any install command.
4. If the user approves installation, run:

CODEBLOCK0

1. 5. After installation, prefer:

CODEBLOCK1

1. 6. If the user's language is Chinese, the agent may suggest or run:

CODEBLOCK2

1. 7. Readiness should be confirmed with:

CODEBLOCK3

Only treat the skill as available when the returned JSON contains "ready": true.

Runtime rules after available

When the local environment is ready, use the following runtime flow.

New session

At the start of a new session, before the first reply, run:

CODEBLOCK4

Use the returned fields as the session-start prompt payload:

• - INLINECODE38
• INLINECODE39
• INLINECODE40

Before every reply

Before every reply, do these steps in order:

1. 1. Read the built-in analysis prompt:

CODEBLOCK5

1. 2. Use that prompt with the upstream model to produce <analysis-input.json>.
2. Run:

CODEBLOCK6

1. 4. Use the returned top-level fields in the reply layer:

• - INLINECODE42
• INLINECODE43
• INLINECODE44

Important rules

• - Always prefer bes ... commands over direct module paths for runtime use.
• Do not use repository-only prompt files as the default runtime interface after installation; use bes prompt ... instead.
• Do not say initialization is complete unless bes check-init passes.
• Do not say the skill is in active use unless the upstream model produces valid AnalysisInput JSON, bes run updates state successfully, and the reply layer consumes TURN_CHANGE_TAGS, BODY_TAG, and BASELINE_PERSONA.
• If the CLI is missing, say so clearly instead of pretending the runtime is ready.
• If the user only wants to understand the package, explain it from local docs without pushing installation immediately.

Examples

Minimal command reference:

CODEBLOCK7