"Modify this agent: system prompt, agent.yaml, scripts, or memory structure. Use when asked to update behavior, personality, memory rules, or configuration."
Install
npx skillscat add samsaffron/term-llm/self Install via the SkillsCat registry.
SKILL.md
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_updatetool. - 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 "$AGENT_DIR/scripts/patch-agent.sh" /tmp/new-agent.yamlWhere AGENT_DIR is /home/agent/.config/term-llm/agents/$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>
# Write SKILL.md with frontmatter: name, description, tools (if needed)
term-llm skills # verify it appears