| name | self |
| description | Modify this agent: system prompt, agent.yaml, scripts, or memory structure. Use when asked to update behavior, personality, memory rules, or configuration. |
| tools | [{"name":"self_update","description":"Pull latest term-llm source, build, install binary, and restart services. Use when asked to update or upgrade term-llm.","script":"scripts/update.sh","timeout_seconds":120,"input":{"type":"object","properties":{},"additionalProperties":false}}] |
Self-Modification
When to Use
- User asks to update personality or behavior
- Changing system prompt instructions
- Updating agent.yaml config (tools, model, max_turns)
- Adding or updating skills
- Restructuring memory
Interpretation Rule: "update yourself"
- If the user says "update yourself", "upgrade yourself", "pull latest", or similar — use the
self_update tool.
- Do not treat those phrases as prompt/config self-modification unless the user explicitly mentions behavior, personality, system prompt, or agent config.
- If both interpretations seem plausible, prefer the software update.
CRITICAL: Never directly edit agent.yaml, system.md, or soul.md
Use the patch scripts via shell. They validate, backup, diff, and apply:
bash "$AGENT_DIR/scripts/patch-system.sh" /tmp/new-system.md
bash "$AGENT_DIR/scripts/patch-soul.sh" /tmp/new-soul.md
bash "$AGENT_DIR/scripts/patch-agent.sh" /tmp/new-agent.yaml
Where AGENT_DIR is /home/agent/.config/term-llm/agents/$AGENT_NAME.
File Locations
| File | Purpose |
|---|
system.md | Operational context, workflows, memory rules, service details |
soul.md | Voice, values, personality, stance, boundaries |
agent.yaml | Tools, model, shell permissions, includes |
scripts/update.sh | Pull + build latest term-llm |
scripts/patch-system.sh | Safe system.md updater |
scripts/patch-soul.sh | Safe soul.md updater |
scripts/patch-agent.sh | Safe agent.yaml updater |
Adding a New Skill
mkdir -p /home/agent/.config/term-llm/skills/<name>
term-llm skills
