Imported from executiveusa/YAPPYVERSE-FACTORY (AGENTS.md). Install upstream with npx skills add executiveusa/YAPPYVERSE-FACTORY. Copyright stays with the author.

AGENTS.md — YAPPYVERSE-FACTORY Agent Roster

The Full Agent Fleet — Topology, Capabilities, Invocation

Authority: EMERALD_TABLETS.md Part V (canonical topology)
Owner: Agent Zero (Claude Code / GitHub Copilot)
Version: 1.0 | March 2026

All agents MUST read EMERALD_TABLETS.md before operating. All commits use ZTE format: [SYNTHIA][BEAD-ID] type: what | LP# | score delta All secrets stay in .env or Infisical. Nothing hardcoded. Ever.

AGENT ZERO

Exclusive authorities:

• Git commits to main
• PR merge decisions (after AI review + CI green)
• Vercel deploy approvals
• PopeBot dispatch (SOP-006)
• Secret rotation (SOP-007)
• Multi-service deploy plans (blast radius > 3)

Invocation: Active by default in this repo. No explicit invocation needed.
Constraints: Never skip quality gates (EMERALD_TABLETS Part VIII). Never print secrets.

JCODEMUNCH

Capabilities:

• list_repos — view all indexed repositories
• index_folder("<path>") — build symbol index of a directory
• search_symbols("<name>") — find any function/class/variable by name
• get_symbol("<file>", "<symbol>") — retrieve exact symbol definition
• get_context_bundle("<file>", "<symbol>") — callers + callees + docs
• get_file_outline("<file>") — structure without content
• invalidate_cache — refresh after large merges

Invocation:

Load jcodemunch-mcp and follow SOP-001 for all code lookups.

Constraints: Index must be valid before symbol search. Invalidate after large merges.

RALPHY

Capabilities:

• Single task execution with --copilot engine
• PRD-driven multi-task with --prd PRD.md
• Parallel branch-per-task with --parallel --branch-per-task --create-pr
• Dry-run preview with --dry-run

Invocation:

ralphy --copilot "task description"
ralphy --copilot --prd PRD.md --parallel --max-parallel 3 --branch-per-task --create-pr

Constraints: Do not modify .ralphy/config.yaml mid-session. Max 3 parallel agents without approval.

AUTONOMOUS_CI_AGENT

6 Phases:

1. Discovery — read state, collect open PRs
2. AI Code Review — OWASP + quality + schema + asset checks
3. CI Green Gate — poll GitHub Actions (10 min timeout)
4. Squash Merge — PUT /pulls/{n}/merge
5. Vercel Deploy — poll deployments, retry up to 5×
6. Live URL Check + Final Export — HTTP 200, write deploy_result.txt

Invocation:

Load E:\YAPPYVERSE-FACTORY\.github\AUTONOMOUS_CI_AGENT.prompt.md
and execute the full 6-phase autonomous CI/CD workflow for all open PRs.

Constraints: Never stop between phases. Never push to main except Phase 4 auto-fix. Never print secrets.

SYNTHIA™ MCP

Tools exposed (when running as MCP server):

• synthia_audit(url) — 12-axis + UDEC score with violations + fixes
• synthia_migrate(wp_xml, routes) → Vercel preview URL + spec file
• synthia_graph_query(cypher) — Neo4j pattern store query
• synthia_score_architecture(description) — Meadows stock/flow/feedback analysis

Skill invocation:

Load .claude/skills/synthia-architecture/SKILL.md and score this architecture.

MCP install:

{
  "mcpServers": {
    "synthia": {
      "command": "npx",
      "args": ["@kupurimedia/synthia-mcp", "proxy"],
      "env": { "SYNTHIA_LICENSE_KEY": "your-license-key" }
    }
  }
}

Constraints: Must have EMERALD_TABLETS.md present (stub skill points to it). Floor 8.5 blocks commits.

POPEBOT FLEET

Dispatch flow:

1. Agent Zero creates job/<task-slug> branch
2. Agent Zero pushes branch → GitHub Action triggers
3. PopeBot Docker container runs Pi coding agent
4. Pi commits + opens PR targeting main
5. Agent Zero processes PR via AUTONOMOUS_CI_AGENT

Direct dispatch:

POST /api/create-job
X-API-Key: <AGENT_GH_TOKEN>
{"job_description": "task description", "priority": "normal"}

Constraints: Agent Zero authority only. Cost guard: $10/job. Cannot merge, rotate secrets, or spawn other bots.

AGENT-BROWSER (E2E)

Capabilities:

• Open URLs and take screenshots
• Test complete user journeys
• Validate database state after form submissions
• Responsive checks across viewports
• Bug hunting across UI, logic, data, and security layers

3-agent parallel research:

• Sub-agent 1: App structure + user journeys + startup commands
• Sub-agent 2: DB schema + data flows + validation queries
• Sub-agent 3: Bug hunting (logic, UI, security, data)

Invocation:

agent-browser open <url>
# Then follow SOP-004 for full 6-phase e2e sweep

Constraints: WSL only on Windows. Screenshots → e2e-screenshots/<journey>/NN-description.png.

AGENT INTERACTION RULES

1. Reading order: EMERALD_TABLETS.md → CLAUDE.md → relevant skill files → task
2. Commit format: [SYNTHIA][BEAD-ID] type: what | LP# | score delta (no exceptions)
3. Blast radius: No single action touches > 3 services without explicit plan + approval
4. Secrets: All in .env or Infisical. Never in code. Never in commits.
5. Quality gate: SYNTHIA™ score ≥ 8.5 before architectural commits. FBK/RSL ≥ 7.0 or redesign first.
6. Cost guard: $10/PopeBot job, $50/day API total. Halt and alert if exceeded.
7. Circuit breaker: Same failure 3× → escalate, do NOT retry same approach.
8. Irreversible actions: Always halt and confirm with user before proceeding.

OPEN BRAIN

Capabilities:

• brain.save_memory(title, content, ...) — persist insights, code, tasks, conversations
• brain.search(query) — ranked full-text search (BM25-style with tsvector weights)
• brain.search_by_vector(embedding, min_similarity) — cosine similarity (384-dim HNSW)
• brain.tag_memory(id, tag) / brain.get_tags(id) — flat tag vocabulary
• brain.link_memories(a, b, relationship) — directed knowledge graph edges
• brain.start_conversation(title) / brain.add_message(conv_id, role, content) — session logging
• brain.set_preference(key, value) / brain.get_preference(key) — agent settings
• brain.create_collection(name) / brain.add_to_collection(coll_id, mem_id) — curated sets
• brain.stats() — quick table counts

Invocation:

from tools.open_brain import OpenBrain
brain = OpenBrain()  # env vars: SUPABASE_SERVICE_ROLE_KEY, DATABASE_URL

Connection method: Supabase pg meta /pg/query API — no direct PostgreSQL port required. Credentials in .env (see .env.template). Schema in db/second_brain_schema.sql.

Constraints: Use service role key. Never expose credentials in commits. Schema is idempotent — re-run db/second_brain_schema.sql to restore after wipe.

AGENTS.md v1.0 | YAPPYVERSE-FACTORY | Agent Zero authority | Kupuri Media™