OwnwareDocs
OwnwareEngine
View source
concept

Example profiles

Ownware bundles a set of ready-made profiles. Some are core (built in); the rest are marketplace profiles — verified examples you install explicitly. Each is just a folder of text files, so the best way to learn the format is to read one. They ship from two places: the core profiles live in packages/cortex/profiles/, and the marketplace profiles in the repo-root profiles/ — so ls profiles/ shows the marketplace set, not the core ones.

Each card below shows the profile's real agent.json settings: its tool preset, security level, and — for the multi-agent ones — the specialist subagents it coordinates.

Core profiles

ownware and ownware-code are the two general-purpose lobby agents. The other three are private helpers (kind: helper) that live inside a parent profile's helpers/ folder — they're spawned by their parent, not run directly. The Access column is the effective tool set after the profile's allow/deny list (not just the raw preset):

Profile What it does Effective access Level
ownware The default general assistant — reads, writes, searches, remembers full (all built-ins) standard
ownware-code Coding partner — edits files, runs commands, reviews diffs full (all built-ins) standard
explore Read-only code explorer (helper) — searches and traces without editing full minus writes/shell/spawn (read-only in practice) standard (zones off)
planner Turns a goal into a step-by-step plan (helper) full minus writes/shell/spawn standard (zones off)
verifier Checks that a change did what it claims (helper) none + readFile, listFiles, glob, grep, shell_execute (runs shell) standard

Marketplace profiles

Installable, domain-specific agents. Each is a complete worked example of the profile format.

Research — ownware-research

Read-only research specialist: searches, reads, analyzes, and reports with file-and-line citations. Never modifies anything.

profiles/ownware-research/agent.json (excerpt)
{
  "name": "ownware-research",
  "model": "anthropic:claude-sonnet-4-6",
  "smallFastModel": "anthropic:claude-haiku-4-5",
  "tools": { "preset": "readonly", "deny": ["shell_execute"] },
  "security": { "level": "strict", "permissionMode": "ask" }
}

The pairing to notice: readonly preset + strict security is how you build an agent that physically cannot change your files, no matter what it's asked. Read its SOUL.md to see how the personality reinforces the same boundary.

A multi-agent legal desk. The lead coordinates four specialist subagents, and ships with reusable skills (review-contract, draft-agreement, due-diligence, legal-memo, …).

profiles/ownware-law/agent.json (excerpt)
{
  "name": "ownware-law",
  "maxTurns": 100,
  "tools": { "preset": "full" },
  "skills": { "dirs": ["skills/"] },
  "subagents": [
    { "name": "researcher", "description": "Finds case law, statutes, precedent" },
    { "name": "analyst",    "description": "Deep-reads contracts; extracts terms & risks" },
    { "name": "drafter",    "description": "Writes contracts, memos, briefs" },
    { "name": "checker",    "description": "Scans against GDPR/HIPAA/SOC2/CCPA" }
  ]
}

This is the shape of a team profile: one lead, several named specialists, plus a skills/ folder of repeatable workflows. See Multi-agent teams for how the coordination works.

Finance — ownware-finance

Senior finance analyst across banking, equity research, PE, wealth, and corp-fin. Builds DCFs, comps, and LBOs; drafts pitchbooks and IC memos. Six subagents: filings-explorer, valuation-builder, earnings-reviewer, market-researcher, diligence-runner, deck-author. Cites every figure to a primary source and refuses to fabricate numbers or give investment advice — the refusals live in its SOUL.md, not in code.

Security — ownware-security

Authorized vulnerability-assessment agent with multi-agent orchestration. Uses the full preset because it needs to run scanners, coordinated by six subagents: recon, vuln-hunter, validator, reporter, fixer, code-reviewer.

Marketing — ownware-marketing

Marketing operator across CRO, copy, SEO, paid, lifecycle, and analytics. Five subagents from audience-researcher to asset-author. Cites audience claims to a source and refuses fabricated testimonials or dark patterns.

Design — ownware-design

A designer who codes: produces landing pages, dashboards, decks, and mobile mockups as HTML. Single-agent, coding preset — a good example of a focused profile with no subagents.

Trading — ownware-trade-coach

Connects to your live broker, watches your fills, and guards against your worst habits. Runs at strict security with permissionMode: ask — anything that touches the account pauses for your approval.

Reading list

The fastest way to internalize the format:

  1. Start with ownware-research — the smallest complete profile.
  2. Then ownware-law — see how subagents and skills are added.
  3. Compare a readonly/strict profile against a full/standard one to feel how the security levels change behavior.

Next steps