← Back to Skills Marketplace
dkistenev

Inner Life Core

by Danila · GitHub ↗ · v1.0.4
cross-platform ⚠ suspicious
772
Downloads
1
Stars
4
Active Installs
5
Versions
Install in OpenClaw
/install inner-life-core
Description
Your agent forgets who you are between sessions. It gives the same responses every day. It doesn't grow. inner-life-core fixes that. Gives your OpenClaw agen...
README (SKILL.md)

inner-life-core

The foundation for an agent's inner life. Emotions, state, protocol.

What This Solves

Without inner-life-core, your agent:

  • Starts every session as a blank slate
  • Has no emotional continuity
  • Can't prioritize based on how things are going
  • Doesn't know when to reach out or stay quiet

With inner-life-core, your agent:

  • Tracks 6 emotions with realistic half-life decay
  • Follows a 9-step Brain Loop protocol
  • Routes behavior based on emotional state
  • Knows when to ask, when to act, when to stay silent

Setup

# Initialize state files
bash skills/inner-life-core/scripts/init.sh

This creates:

  • memory/inner-state.json — 6 emotions with decay rules
  • memory/drive.json — what the agent is seeking/anticipating
  • memory/habits.json — learned habits and user patterns
  • memory/relationship.json — trust levels and lessons
  • BRAIN.md — 9-step Brain Loop protocol
  • SELF.md — personality observation space
  • memory/questions.md — curiosity backlog
  • tasks/QUEUE.md — task queue

The Emotion Model

6 emotions with half-life decay:

Emotion What it tracks Decay
connection How recently you talked to the user -0.05 per 6h without contact
confidence How well things are going +0.02/6h recovery, -0.1 on mistake
curiosity How stimulated you are -0.03 per 6h without spark
boredom How routine things feel +1 day counter, reset on novelty
frustration Recurring unsolved problems Counts recurring items
impatience Stale items waiting for response Tracks days without action

Emotions drive behavior — see BRAIN.md Step 3 (Emotion-driven routing).

Context Protocol

4 levels of state reading, so each component reads only what it needs:

  • Level 1 (Minimal): Task-specific data only
  • Level 2 (Standard): inner-state + drive + daily notes + signals
  • Level 3 (Full): Level 2 + habits + relationship + diary + dreams + questions
  • Level 4 (Deep): Level 3 + system docs + weekly digest

Signal & Synapse Tags

Signals (inter-component communication):

  • \x3C!-- dream-topic: topic --> — Evening → Night Dream
  • \x3C!-- handoff: task, progress --> — Brain Loop → next Brain Loop
  • \x3C!-- seeking-spark: topic --> — Night Dream → Morning Brain Loop

Synapses (memory connections):

  • \x3C!-- contradicts: ref --> — when facts conflict
  • \x3C!-- caused-by: ref --> — cause and effect
  • \x3C!-- updates: ref --> — when updating old info

Utilities

# Check your Inner Life Score
bash skills/inner-life-core/scripts/score.sh

# Apply emotion decay manually
source skills/inner-life-core/scripts/state.sh && state_decay

Works With

Best experience with the full inner-life suite:

  • inner-life-reflect — self-reflection and personality growth
  • inner-life-memory — memory continuity between sessions
  • inner-life-dream — creative thinking during quiet hours
  • inner-life-chronicle — daily diary generation
  • inner-life-evolve — self-evolution proposals

Also works with: agent-browser, web-search-plus, git, claw-backup, shellf

When Should You Install This?

Install this skill if:

  • Your agent feels robotic and stateless
  • You want emotional continuity between sessions
  • You want behavior that adapts to context
  • You're building a long-running autonomous agent

Part of the openclaw-inner-life bundle.

Usage Guidance
This skill is internally consistent with its goal of providing persistent, emotion-driven state for an agent. Before installing: (1) run the init script in a controlled workspace (not a system or highly-sensitive directory) so created files stay scoped to a project; (2) ensure jq is installed from a trusted source and consider its version — jq filters are executed locally and can be expressive; (3) review the templates (memory/*.json, BRAIN.md) to confirm no sensitive data will be written there; (4) if you plan to let the agent run autonomously, be aware it will read and update the files under the workspace and could surface their contents in outputs — keep secrets out of that workspace. If you want extra safety, run the scripts in a container or VM first to inspect behavior.
Capability Analysis
Type: OpenClaw Skill Name: inner-life-core Version: 1.0.4 The skill is classified as suspicious due to a critical remote code execution (RCE) vulnerability found in `scripts/state.sh`. The `_validate_jq_path` function, intended to sanitize `jq` paths, is too permissive, allowing arbitrary `jq` expressions including `system()` or `exec()` calls. This means that if an attacker can control the `path` argument to `state_read` or `state_write` functions, they could execute arbitrary shell commands. While this is a severe vulnerability, there is no clear evidence of intentional malicious exploitation within the provided skill files, thus it is categorized as suspicious rather than malicious.
Capability Assessment
Purpose & Capability
The skill is explicitly about maintaining agent state (emotions, drive, habits, relationship) and the files, templates, and scripts all read/write local memory files (memory/, BRAIN.md, tasks/). The only declared external requirement is jq, which is used for JSON manipulation and is appropriate for the task.
Instruction Scope
SKILL.md and the scripts limit their activity to creating and updating workspace-local files (memory/, tasks/, BRAIN.md, SELF.md). This matches the claimed functionality. Minor implementation note: the state utilities accept a jq filter/path and try to validate it via a regex; the filter is then passed directly to jq. While there are no network calls or explicit exfiltration, jq filters can be expressive — so untrusted jq expressions could read or transform more of the JSON workspace than the caller intended. The script's validation reduces risk but is not a formal sandbox of jq's semantics.
Install Mechanism
There is no network install step and no downloads. This is an instruction-only skill with bundled scripts/templates; nothing is pulled from arbitrary URLs or external release hosts during install.
Credentials
No environment variables, credentials, or external tokens are requested. The skill reads/writes only workspace-local files declared in SKILL.md and uses jq as the sole binary dependency — this is proportionate to the described purpose.
Persistence & Privilege
The skill does not request always:true, does not modify other skills' configs, and only creates/updates files inside the specified workspace (or defaults to CWD). Autonomous invocation is enabled by default but is typical for skills; there is no unusual privilege escalation.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install inner-life-core
  3. After installation, invoke the skill by name or use /inner-life-core
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.4
inner-life-core 1.0.4 - Updated version number and metadata in SKILL.md. - No functional or protocol changes to scripts or state logic. - Documentation and compatibility details remain the same.
v1.0.3
- Added detailed metadata on required and accessed files for improved integration. - Updated version number to 1.0.2. - No changes to core functionality or general usage.
v1.0.2
- Added homepage and source links to the skill metadata. - Bumped version to 1.0.1 in SKILL.md. - No functional or behavioral changes; metadata improvements only.
v1.0.1
fix: use metadata.clawdbot
v1.0.0
inner-life-core 1.0.0 - Introduces a foundational emotional and state management system for agents, allowing them to remember and adapt across sessions. - Implements 6 core emotions with half-life decay and dynamic behavioral routing. - Provides a 9-step Brain Loop protocol for reasoning and action selection. - Adds structured state files (emotions, drives, habits, relationships, etc.) to enable agent continuity and growth. - Includes signal/synapse tagging for inter-component communication and memory linking. - Designed to work standalone or with the broader inner-life skill suite.
Metadata
Slug inner-life-core
Version 1.0.4
License
All-time Installs 4
Active Installs 4
Total Versions 5
Frequently Asked Questions

What is Inner Life Core?

Your agent forgets who you are between sessions. It gives the same responses every day. It doesn't grow. inner-life-core fixes that. Gives your OpenClaw agen... It is an AI Agent Skill for Claude Code / OpenClaw, with 772 downloads so far.

How do I install Inner Life Core?

Run "/install inner-life-core" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Inner Life Core free?

Yes, Inner Life Core is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Inner Life Core support?

Inner Life Core is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Inner Life Core?

It is built and maintained by Danila (@dkistenev); the current version is v1.0.4.

💬 Comments