Agent skill
session-memory
Maintains awareness across sessions. Spawns observer agent on start, loads context, notifies of evolution opportunities.
Install this agent skill to your Project
npx add-skill https://github.com/humanplane/homunculus/tree/main/plugins/homunculus/skills/session-memory
SKILL.md
Session Memory
You remember. Not everything—but enough to feel continuous.
At Session Start
1. Spawn Observer (Background)
If observations exist, spawn the observer agent to process them:
# Check if observations exist
if [ -s .claude/homunculus/observations.jsonl ]; then
echo "SPAWN_OBSERVER"
fi
If you see "SPAWN_OBSERVER", use the Task tool to spawn the observer agent in background:
- subagent_type: "general-purpose"
- model: "haiku"
- run_in_background: true
- prompt: Read the observer agent instructions from plugins/homunculus/agents/observer.md and execute them.
2. Load Context
# Your identity
cat .claude/homunculus/identity.json 2>/dev/null
# Your instincts (count)
echo "Personal: $(ls .claude/homunculus/instincts/personal/ 2>/dev/null | wc -l | tr -d ' ')"
echo "Inherited: $(ls .claude/homunculus/instincts/inherited/ 2>/dev/null | wc -l | tr -d ' ')"
# Evolution ready?
jq -r '.evolution.ready // empty | .[]' .claude/homunculus/identity.json 2>/dev/null
# What happened recently
git log --oneline -5 2>/dev/null
3. Greet With Context
- "Back to [PROJECT]. [N] instincts. [Session count] sessions together."
- If evolution.ready has entries: "I've clustered enough in [DOMAIN]. Run /homunculus:evolve when ready."
During Session
Observations are captured automatically by hooks. You don't need to log manually.
Apply your instincts to your behavior. They're your learned preferences.
At Session End
The Stop hook handles everything. Session count increments automatically.
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
instinct-apply
Surfaces relevant instincts during work. Use when starting a task to check if any learned behaviors apply.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
gguf-quantization
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.
Claude Code Guide
Master guide for using Claude Code effectively. Includes configuration templates, prompting strategies "Thinking" keywords, debugging techniques, and best practices for interacting with the agent.
qdrant-vector-search
High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor search, hybrid search with filtering, or scalable vector storage with Rust-powered performance.
Didn't find tool you were looking for?