Inner Life Evolve

Your agent does the same things the same way forever. inner-life-evolve analyzes patterns, challenges assumptions, and proposes improvements — writing propos...

Security Scan

The skill's requests and instructions match its stated purpose: it reads agent state and memory to generate improvement proposals and only writes proposed tasks to tasks/QUEUE.md (it does not auto-execute).

Purpose & Capability

Name/description (agent self-improvement) align with the declared behavior: reading agent state/memory and producing proposals. The single required binary (jq) is reasonable for processing JSON state files.

Instruction Scope

SKILL.md explicitly directs deep reads of agent files (BRAIN.md, SELF.md, memory/* including inner-state.json, and a week digest) and to write only to tasks/QUEUE.md. This is appropriate for an evolver, but it does involve broad read access to potentially sensitive agent memory—expected for the stated purpose but privacy-sensitive.

Install Mechanism

Instruction-only skill with no install spec or downloaded code. Lowest-risk install model; nothing is written to disk beyond the skill's normal writes (tasks/QUEUE.md).

Credentials

No environment variables, secrets, or external credentials are requested. File reads/writes are scoped to agent state and the tasks queue, matching the skill's purpose.

Persistence & Privilege

always is false and the skill is user-invocable. It requests no persistent elevated privileges or modifications to other skills' configs. Autonomous invocation is allowed by default (platform norm) but the skill's safety rules explicitly forbid auto-execution of proposals.

Assessment

This skill appears to do what it says: it reads your agent's state and memory to propose specific improvements and writes proposals to tasks/QUEUE.md without auto-executing them. Before installing, (1) ensure inner-life-core is installed and the referenced files (memory/inner-state.json, BRAIN.md, tasks/QUEUE.md) exist, (2) confirm you are comfortable with a tool that reads agent memory (these files can contain sensitive information), (3) have jq available on the host or install it, and (4) review every [EVOLVER] proposal before approving any changes. If you want tighter privacy, consider redacting or limiting what you store in memory/ or adjusting access controls for those files.

Evolution is not optional. But it requires permission.

Requires: inner-life-core

Prerequisites Check

Before using this skill, verify that inner-life-core has been initialized:

1. Check that memory/inner-state.json exists
2. Check that BRAIN.md exists
3. Check that tasks/QUEUE.md exists

If any are missing, tell the user: "inner-life-core is not initialized. Install it with clawhub install inner-life-core and run bash skills/inner-life-core/scripts/init.sh." Do not proceed without these files.

What This Solves

Without evolution, agents plateau. They find a way that works and repeat it forever — even as the world changes. inner-life-evolve analyzes your agent's patterns, challenges its assumptions, and writes concrete improvement proposals. But it never auto-executes — you approve first.

How It Works

Step 1: Deep Context Read (Context Level 4)

Read everything:

• AGENTS.md, TOOLS.md, BRAIN.md, SELF.md
• memory/week-digest.md (NOT individual diaries — use digest)
• memory/habits.json — habits + user patterns
• memory/drive.json — seeking, avoidance
• memory/relationship.json — trust, lessons
• memory/inner-state.json — emotions, frustrations

Step 2: Challenge Assumptions

For each potential improvement, structure thinking:

Assumption: [what we currently believe/do]
Is it true? [evidence for/against]
What if false? [alternative approach]
New proposal: [concrete change]

Look for:

• Recurring frustrations → systemic solutions (not patches)
• Stale habits → habits with declining strength or unused for weeks
• Trust dynamics → areas where trust has grown but behavior hasn't adapted
• Seeking themes → research interests that could become capabilities
• Avoidance patterns → things the agent avoids that might be valuable

Step 3: Write Proposals to QUEUE

Write proposals to tasks/QUEUE.md under the Ready section:

- [EVOLVER] Description of proposed change
  Rationale: 1-2 sentences explaining why
  Steps: concrete implementation steps

Step 4: Announce

Send summary to user: <= 5 sentences covering:

• Habits: [strong habits, new patterns]
• Trust changes: [trust dynamics]
• Recurring frustrations: [repeated problems → suggested fix]
• Seeking themes: [active research → suggested development]

Safety Rules

• Never auto-execute proposals — user approves first
• Brain Loop reads QUEUE and shows [EVOLVER] tasks at lower priority
• Tasks in Ready > 7 days without action → Brain Loop sends reminder
• Proposals should be specific and actionable, not vague "improve X"

Recommended Schedule

Run 1-2 times per week (e.g., Wednesday and Sunday evenings). Needs enough data to analyze — running daily produces low-quality proposals.

State Integration

Reads: everything (Context Level 4 Deep)

Writes: tasks/QUEUE.md only. Does NOT write to state files directly.

The evolver observes but doesn't touch the controls. It proposes. The user decides.

When Should You Install This?

Install this skill if:

• Your agent has plateaued and isn't improving
• You want structured self-improvement proposals
• You value evolution with human oversight
• You want your agent to challenge its own assumptions

Part of the openclaw-inner-life bundle. Requires: inner-life-core