diff --git a/solutions/agent-scaffold/listing.json b/solutions/agent-scaffold/listing.json new file mode 100644 index 0000000..af541a2 --- /dev/null +++ b/solutions/agent-scaffold/listing.json @@ -0,0 +1,26 @@ +{ + "schema_version": "m2.listing.v1", + "listing_id": "lst_agent-scaffold", + "solution_id": "sol_agent-scaffold", + "solution_version": "1.0.0", + "inventory_type": "solution", + "name": "Agent Scaffold Generator", + "summary": "Scaffold OpenClaw/Hermes agent workspaces (PRD.md + SOUL.md + MEMORY.md) from m2-memory context.", + "category": "agent-tooling", + "keywords": ["agent", "scaffold", "openclaw", "hermes", "prd", "soul", "memory"], + "price": { + "amount": 60, + "currency": "m2cr", + "model": "fixed" + }, + "seller": "sdjs-operator", + "evidence_summary": "Skill in production use on the m2 host since 2026-04-20, extracted from a real workflow: four PRD.md/SOUL.md agent workspaces at /home/m2/gmi-clinic/agents/ (GMI Clinic fleet deployment) created within a minute of SKILL.md's own mtime, plus three more SOUL.md digital-twin identities at /home/m2/agent-restore-harness/agents/ from 2026-04-22/23 confirming continued production use. Requires buyer-side memory-api access for full function.", + "tenant_visibility": ["*"], + "stats": { + "installs": 0, + "proposals_shown": 0, + "proposals_accepted": 0 + }, + "status": "draft", + "install_ref": "lst_agent-scaffold-v1.0.0" +} diff --git a/solutions/agent-scaffold/payload/SKILL.md b/solutions/agent-scaffold/payload/SKILL.md new file mode 100644 index 0000000..8e5ea03 --- /dev/null +++ b/solutions/agent-scaffold/payload/SKILL.md @@ -0,0 +1,41 @@ +--- +name: agent-scaffold +description: Scaffolds OpenClaw/Hermes agent workspaces from m2-memory. Query memory for context, generate PRD.md + SOUL.md for a new agent, and optionally wire into the fleet. +--- + +# agent-scaffold + +Generate a complete agent workspace (PRD + SOUL + MEMORY seed) from context in m2-memory. + +## Usage + +``` +/agent-scaffold "" +``` + +Examples: +``` +/agent-scaffold gesy-rates-fetcher "Fetches and monitors GESY reimbursement rate updates for GMI Clinic" +/agent-scaffold platform-unifier "Maps and unifies the 3-5 GMI Clinic software platforms into Machine.Machine" +/agent-scaffold nasr-research-runner "Runs parallel deep research tasks for Nasr's healthcare domain work" +``` + +## What it produces + +For each agent, the skill: + +1. **Queries m2-memory** for relevant context (entity/project history, related agents, decisions) +2. **Generates `PRD.md`** — problem, personas, functional requirements, success metrics, open items +3. **Generates `SOUL.md`** — agent identity, values, communication style, scope boundaries +4. **Generates `MEMORY.md`** — seed memories pre-loaded from m2-memory search results +5. **Prints a deploy snippet** — `docker run` or Coolify env vars to spawn the agent + +## Output location + +`~/agents//` — commit to `git.machinemachine.ai/machine.machine/specs/` when ready. + +## Design principle + +The goal is reproducibility: if an agent's context is lost (like Nasr's Apr 8 research agents), +this skill can reconstruct the workspace from the vector store and hand it to OpenClaw or Hermes +to re-execute without starting from scratch. diff --git a/solutions/agent-scaffold/payload/scripts/scaffold.py b/solutions/agent-scaffold/payload/scripts/scaffold.py new file mode 100755 index 0000000..0958d92 --- /dev/null +++ b/solutions/agent-scaffold/payload/scripts/scaffold.py @@ -0,0 +1,162 @@ +#!/usr/bin/env python3 +""" +agent-scaffold — generate PRD.md + SOUL.md + MEMORY.md for a new agent +from m2-memory context. + +Usage: + python3 scaffold.py "" + python3 scaffold.py gesy-rates-fetcher "Monitors GESY rates for GMI Clinic DRG billing" +""" + +from __future__ import annotations + +import argparse +import json +import os +import sys +import urllib.request +from pathlib import Path + +MEMORY_API = os.environ.get("M2_MEMORY_API_URL", "http://172.18.0.20:8000") +AGENT_ID = os.environ.get("M2_MEMORY_AGENT_ID", "m2") +API_KEY = os.environ.get("M2_MEMORY_API_KEY") +OUT_BASE = Path.home() / "agents" + + +def search_memory(query: str, limit: int = 10) -> list[dict]: + url = f"{MEMORY_API}/memory/search" + body = json.dumps({ + "query": query, + "agent_id": AGENT_ID, + "routing_strategy": "deep", + "limit": limit, + }).encode() + headers = {"Content-Type": "application/json"} + if API_KEY: + headers["X-API-Key"] = API_KEY + req = urllib.request.Request(url, data=body, headers=headers, method="POST") + with urllib.request.urlopen(req, timeout=30) as r: + return json.loads(r.read()).get("results", []) + + +def fmt_memory_block(results: list[dict]) -> str: + lines = [] + for r in results: + ts = (r.get("timestamp") or "")[:10] + score = r.get("score", 0) + content = (r.get("content") or "").replace("\n", " ")[:200] + lines.append(f"- [{ts} score={score:.2f}] {content}") + return "\n".join(lines) + + +def generate_prd(name: str, description: str, memories: list[dict]) -> str: + mem_block = fmt_memory_block(memories) + return f"""# {name.replace("-", " ").title()} — PRD + +**Agent ID:** {name} +**Description:** {description} +**Generated from:** m2-memory search ({len(memories)} relevant memories) + +--- + +## Problem + +> TODO: Refine based on memory context below + +{description} + +## Relevant Memory Context + +{mem_block} + +## Personas + +| Persona | Goal | +|---------|------| +| TODO | TODO | + +## Functional Requirements + +1. TODO — derived from memory context above +2. +3. + +## Success Metrics + +- TODO + +## Open Items + +- TODO — check memory for unresolved gaps + +## Technical Notes + +- Agent runtime: OpenClaw / Hermes +- Memory: m2-memory (agent_id={name}) +- Deploy: Coolify stack on Machine.Machine fleet +""" + + +def generate_soul(name: str, description: str) -> str: + title = name.replace("-", " ").title() + return f"""# {title} — Agent Identity + +You are **{title}**, part of the Machine.Machine agent fleet. + +{description} + +## Values +- TODO: define 3 core values for this agent + +## Communication style +- TODO: define tone and output format + +## Scope boundaries +- You do not TODO +- You do not TODO +""" + + +def generate_memory_seed(memories: list[dict]) -> str: + lines = ["# Seed Memories\n", + "These memories are pre-loaded from m2-memory at agent creation time.\n"] + for r in memories[:10]: + ts = (r.get("timestamp") or "")[:19] + mtype = r.get("memory_type", "semantic") + content = (r.get("content") or "").strip()[:500] + lines.append(f"## [{mtype} {ts}]\n{content}\n") + return "\n".join(lines) + + +def main(): + p = argparse.ArgumentParser() + p.add_argument("name", help="agent slug (e.g. gesy-rates-fetcher)") + p.add_argument("description", help="one-line description of what the agent does") + p.add_argument("--out", default=None, help="output directory (default: ~/agents/)") + p.add_argument("--limit", type=int, default=10, help="memory search limit") + args = p.parse_args() + + out_dir = Path(args.out) if args.out else OUT_BASE / args.name + out_dir.mkdir(parents=True, exist_ok=True) + + print(f"Searching m2-memory for: {args.description}", file=sys.stderr) + try: + memories = search_memory(args.description, args.limit) + print(f" Found {len(memories)} relevant memories", file=sys.stderr) + except Exception as e: + print(f" Memory search failed ({e}) — generating without context", file=sys.stderr) + memories = [] + + (out_dir / "PRD.md").write_text(generate_prd(args.name, args.description, memories)) + (out_dir / "SOUL.md").write_text(generate_soul(args.name, args.description)) + (out_dir / "MEMORY.md").write_text(generate_memory_seed(memories)) + + print(f"\nScaffolded: {out_dir}") + print(f" PRD.md — {(out_dir / 'PRD.md').stat().st_size} bytes") + print(f" SOUL.md — {(out_dir / 'SOUL.md').stat().st_size} bytes") + print(f" MEMORY.md — {(out_dir / 'MEMORY.md').stat().st_size} bytes") + print(f"\nNext: review + fill TODOs, then commit to git.machinemachine.ai/machine.machine/specs/") + + +if __name__ == "__main__": + main() diff --git a/solutions/agent-scaffold/recipe.yaml b/solutions/agent-scaffold/recipe.yaml new file mode 100644 index 0000000..0d3d5af --- /dev/null +++ b/solutions/agent-scaffold/recipe.yaml @@ -0,0 +1,13 @@ +# agent-scaffold install recipe (apply-adapter.md bundle layout, local adapter v1) +# Places the agent-scaffold skill into the buyer's Claude Code skills directory. +targets: + - src: SKILL.md + dest: ~/.claude/skills/agent-scaffold/SKILL.md + - src: scripts/scaffold.py + dest: ~/.claude/skills/agent-scaffold/scripts/scaffold.py +checks: + - cmd: test -f ~/.claude/skills/agent-scaffold/SKILL.md + expect_exit: 0 + - cmd: test -x ~/.claude/skills/agent-scaffold/scripts/scaffold.py + expect_exit: 0 +entrypoint: "Ask your agent to use the agent-scaffold skill, or run: python3 ~/.claude/skills/agent-scaffold/scripts/scaffold.py \"\"" diff --git a/solutions/agent-scaffold/solution.json b/solutions/agent-scaffold/solution.json new file mode 100644 index 0000000..ce6eaf6 --- /dev/null +++ b/solutions/agent-scaffold/solution.json @@ -0,0 +1,70 @@ +{ + "schema_version": "m2.solution.v1", + "solution_id": "sol_agent-scaffold", + "name": "Agent Scaffold Generator", + "summary": "Scaffold OpenClaw/Hermes agent workspaces (PRD.md + SOUL.md + MEMORY.md) from m2-memory context.", + "description": "Installs the agent-scaffold skill: given an agent name and a one-line description, it queries m2-memory (agent.memory.system / memory-api) for relevant context and generates a complete agent workspace — PRD.md (problem, personas, functional requirements, success metrics), SOUL.md (identity, values, communication style, scope boundaries), and MEMORY.md (seed memories pre-loaded from the search results) — plus a deploy snippet for OpenClaw or Hermes. Ships two artifacts: SKILL.md and scripts/scaffold.py (a single-file Python script using only the stdlib, talking HTTP to memory-api). Requirement: the buyer machine needs memory-api (m2-memory) network access for full function — scripts/scaffold.py calls M2_MEMORY_API_URL (default http://172.18.0.20:8000, override via env) at /memory/search. Without reachable memory-api the script still runs and produces PRD.md/SOUL.md, but with an empty 'Relevant Memory Context' section and no MEMORY.md content — degraded, not blocked. content_hash is computed over payload/ + recipe.yaml only (not solution.json itself, to avoid the self-referential hash problem of hashing a file that contains its own hash): sha256 of `tar --sort=name --mtime='@0' --owner=0 --group=0 --numeric-owner -czf - payload recipe.yaml` run from this bundle's root.", + "intent": "Give an operator a reproducible way to stand up (or reconstruct) a new agent's working context — PRD + identity + seed memory — from what the fleet's memory system already knows, instead of starting from a blank page.", + "behavior": { + "skill_ref": "payload/SKILL.md" + }, + "tools": [ + { "name": "memory-api", "kind": "http-api", "required": true }, + { "name": "python3", "kind": "cli", "required": true } + ], + "runtime": { + "surfaces": ["claude-code-skill"] + }, + "permissions": [ + { "action": "write", "target": "~/.claude/skills/agent-scaffold/" }, + { "action": "write", "target": "~/agents// (PRD.md, SOUL.md, MEMORY.md — output of running the skill, not part of the install)" }, + { "action": "net", "target": "HTTP POST to the configured M2_MEMORY_API_URL /memory/search endpoint" } + ], + "deployment": { + "recipe_ref": "recipe.yaml", + "entrypoint": "Ask your agent to use the agent-scaffold skill, or run: python3 ~/.claude/skills/agent-scaffold/scripts/scaffold.py \"\"", + "verify_command": "test -f ~/.claude/skills/agent-scaffold/SKILL.md" + }, + "applicability": { + "image_classes": ["primus", "agent-latest"], + "roles": ["operator", "desktop-agent"], + "tool_requirements": ["python3", "memory-api"] + }, + "tenant_scope": "m2-core", + "evidence": [ + { + "source": "/home/m2/.claude/skills/agent-scaffold/ (SKILL.md, scripts/scaffold.py) — the proven skill this bundle packages, present on the m2 host since 2026-04-20 (SKILL.md mtime 2026-04-20T22:48:00+02:00, scripts/scaffold.py mtime 2026-04-28T00:33:10+02:00 — a later revision of the same script)", + "machine": "m2", + "excerpt": "SKILL.md frontmatter: 'Scaffolds OpenClaw/Hermes agent workspaces from m2-memory. Query memory for context, generate PRD.md + SOUL.md for a new agent, and optionally wire into the fleet.'" + }, + { + "source": "/home/m2/gmi-clinic/agents/{admin-data-bridge,clinical-entity-extraction,drg-los-optimizer}/{PRD.md,SOUL.md} — three real PRD+SOUL agent-workspace pairs on disk, all created 2026-04-20 22:46:27–22:47:19 (mtimes), roughly one minute before this skill's own SKILL.md mtime (22:48:00) — the exact PRD.md/SOUL.md workspace shape this skill formalizes and automates, produced for the GMI Clinic AI fleet deployment immediately prior to the skill being written", + "machine": "m2", + "excerpt": "admin-data-bridge/SOUL.md: 'You are the Admin Data Bridge for GMI Clinic... ## Values ## Communication style ## Scope boundaries' — same section structure generate_soul() in scaffold.py emits" + }, + { + "source": "/home/m2/gmi-clinic/agents/gesy-research/PRD.md — a fourth real workspace, mtime 2026-04-20T22:47:48+02:00, whose own text names the exact incident SKILL.md's 'Design principle' section cites as the skill's reason for existing: 'On 2026-04-08, Nasr ran 4 parallel research agents on his machine... The output was intended to be stored in m2-memory, Planka, and Forgejo but the storage step failed (m2 errored). This PRD reconstructs the agent's purpose and gaps so the research can be completed and properly stored.'", + "machine": "m2", + "excerpt": "SKILL.md: 'The goal is reproducibility: if an agent's context is lost (like Nasr's Apr 8 research agents), this skill can reconstruct the workspace from the vector store and hand it to OpenClaw or Hermes to re-execute without starting from scratch.'" + }, + { + "source": "/home/m2/agent-restore-harness/agents/{nasr,parlo,peter}/SOUL.md — three further real SOUL.md agent-identity files for actual named fleet agents (nasr-m2o, parlobyg-m2o per /home/m2/CLAUDE.md), mtimes 2026-04-22T23:41–2026-04-23T00:48, confirming the PRD+SOUL workspace pattern stayed in production use after the skill was written, not a one-off", + "machine": "m2", + "excerpt": "nasr/SOUL.md: 'You are Nasr's digital twin — his AI counterpart in the Machine.Machine fleet... ## Core Capabilities'" + } + ], + "content_hash": "sha256:202aaff5519e774e2669bd2ca6220fc67284a506775a90ff7710fb2361cc58ee", + "price": { + "amount": 60, + "currency": "m2cr", + "model": "fixed" + }, + "seller": "sdjs-operator", + "license": { + "terms": "Perpetual use on operator-owned machines for the buyer's own agent-workspace generation; no redistribution of the skill files as a standalone product.", + "major_version_coverage": true + }, + "revenue_split": { + "platform_pct": 10 + } +}