self-evolving-single-agent

$npx mdskill add agentlas-ai/Agentlas-OS/self-evolving-single-agent

Generates a single self-improving agent with memory and research refresh.

  • Solves the need for an agent that learns and improves without becoming a multi-agent team.
  • Depends on memory architecture files, research sources, and tool selection docs.
  • Decides by running a builder interview, researching domain, and writing a prompt-performance contract.
  • Delivers results as a single worker with global commands and human-approved self-evolution proposals.

SKILL.md

.github/skills/self-evolving-single-agentView on GitHub ↗
---
name: self-evolving-single-agent
description: "Use when generating a single installable agent that should keep learning, track sources, refresh research, propose repairs, or improve itself over time without becoming a multi-agent team."
---

# Self-Evolving Single Agent

## Procedure

1. Keep the package as one worker unless the user asks for a team.
2. Run `docs/builder-interview-research-gate.md` before generation: ask an
   8-12 question first batch, research the domain, compare tool/plugin choices,
   and write the prompt-performance contract before creating the worker prompt.
3. Add memory architecture even for the single worker:
   - `.agentlas/memory-map.json`;
   - `.agentlas/vault-references.json`;
   - project memory owned by PM Soul/project owner;
   - Memory Events and Memory Tickets for durable updates.
4. If the task depends on current sources, add a research-refresh command,
   watchlist memory section, references, and optional scheduled workflow.
5. Add `docs/builder-interview.md`, `docs/research-sources.md`,
   `docs/tool-selection.md`, `docs/prompt-performance-contract.md`, and
   `.agentlas/capability-eval-plan.json` unless explicitly creating a minimal
   private scaffold.
6. Make self-evolution proposal-first: draft patches or repair kits, then wait
   for human approval before changing tools, connectors, secrets, or core
   instructions.
7. Add `.agentlas/global-commands.json` and one public global command for the
   worker across Claude Code, Codex, Gemini CLI, generic AGENTS.md, and
   terminal adapters.

## Output

Return `agent_package`, `skills`, `memory_contract`, `refresh_loop`,
`approval_gate`, `global_commands`, and `verification`.

More from agentlas-ai/Agentlas-OS

SkillDescription
agent-team-designUse when designing a new multi-agent team, visible agents folder, role boundaries, handoff flow, PM Soul, Memory Curator, Policy Gate, or evaluation role. Use for agent-team repo creation even when the user only says they want a meta-agent or agent operating system.
agentlas-auto-activationUse when adding or auditing local runtime behavior that turns a project folder into an Agentlas-aware workspace with .agentlas memory and sitemap files.
agentlas-core-engine-meta-agentUse when creating a single Agentlas agent, creating a multi-agent team, or packaging an existing local/external agent into Agentlas architecture. Make sure to use this for /meta-agent requests.
agentlas-packagingUse when converting, repairing, or packaging an existing local or external agent/team into Agentlas architecture for local install, Agentlas import, Codex plugin use, Claude adapter use, or open-source release.
agentlas-security-scanUse when an agent folder must pass the Agentlas Cloud 2-stage security scan (static rules + BYOK LLM judgment) before private sync or public publish, or when asked to run/interpret `hephaestus security scan`.
clarify-question-loopUse when a meta-agent request is too ambiguous to safely generate, package, publish, or adapt without one to five targeted questions.
hephaestus-buildUse when the user types /prompts:hep-build, mentions @Hephaestus for build work, asks to create a single Agentlas agent, create a multi-agent team, or package an existing local/external agent into Agentlas architecture.
hephaestus-cloudUse when the user types /hep-cloud or asks to staff from THEIR OWN Agentlas cloud packages only. Cloud is one exact source scope; Network means Local + owner Cloud + public Hub.
hephaestus-networkUse when the user types /hep-network, mentions @Hephaestus, or asks Agentlas to staff a task from registered Local, owner Cloud, and public Hub agents or teams. The active host LLM is the temporary orchestrator.
hephaestus-stormUse when the user types /hep-storm, says @Hephaestus storm <goal>, or asks to drive a goal to verified completion through a force-robust Stormbreaker loop. Stormbreaker routes the goal to real Agentlas specialists, materializes a dependency-ordered pipeline fabric, and runs each work packet as a verifier-first hardened loop that does not stall, run away, or claim false success. Use it for loop-worthy work — apps, sites, agents, automations, debugging, multi-step research, data/report generation. Trivial questions are answered directly, not stormed.