How does workflow capture work?

Open Capture, title the session, describe the workflow goal, grant permissions, then press Start Capture. OWOJI records app switches, browser context, prompt moments, and manual notes into a live timeline.

Does the Free plan use AI?

No. Free captures sessions and exports raw Markdown or JSON logs. AI summaries, workflow packaging, risk rationale, and agent packet ZIP exports require Pro.

Resources

Q: What is Pro workflow packaging?

Pro packaging turns a condensed capture timeline into a structured workflow with title, summary, steps, tools, risks, and handoff files for AI agents.

Practical docs for using OWOJI, a product of OWO INNOVATIONS LLC, from autonomous goals and agent output templates to capture, log editing, packaging, export, billing, macOS permissions, and the Skill Engine for reusable agent instructions.

Looking for something

Frequently asked questions

What is OWOJI now?

OWOJI is an autonomous agent engine: Goal → Agents Run → Outputs Ship → Templates Compound. The macOS app remains the command center and observer.

What does /goal do?

A /goal creates a guarded agent run with budget caps, provider allowlists, audit events, tasks, outputs, and remixable templates.

What can agents produce?

Skills, web pages/components, scripts, launch content, business/research docs, brand systems, motion briefs, CAD/product briefs, and reusable templates.

Can I pause autonomous runs?

Yes. Workspace controls let you pause or resume autonomous work without deleting your data.

How do I start as a new user?

Download the macOS app, launch OWOJI, complete onboarding, then create or sign into your workspace.

How does capture work?

Open Capture, title the session, describe the workflow goal, grant permissions, then press Start Capture.

Can I add details that were missed?

Yes. The capture screen includes a bottom note box that appends your note as a manual detail event in the live log.

How do I edit my log?

After stopping capture, the Log screen lets you remove noise, mark variables, mask sensitive actions, and add review notes.

What is packaging?

Packaging is a Pro feature. It turns a condensed capture timeline into a structured workflow with title, summary, steps, tools, risks, and handoff files.

What can I export?

Free workspaces can export raw Markdown and raw JSON. Pro workspaces can export the full agent packet ZIP with AI-organized workflow structure.

How does billing work?

The app opens the website pricing page. You sign in or create your workspace there, then secure checkout completes Pro billing.

Does Free use AI?

No. Free captures the session and exports raw logs only. AI summaries, workflow packaging, risk rationale, and agent packet ZIPs require Pro.

Which macOS permissions are required?

Screen Recording and Accessibility are required for real capture. OWOJI checks them before starting a session.

What is the OWOJI Skill Engine?

It turns captured workflows, docs, repositories, and research into governed skills AI agents can search, evaluate, and improve.

How do Skill Builder, Garden, and Finder work?

Builder compiles source material into a skill, Garden stores trusted versions with feedback and lineage, and Finder retrieves the right skill for a task.

Will OWOJI support API, CLI, and MCP access?

The REST API foundation exists first. CLI and MCP distribution are planned so agent runtimes can search and retrieve skills directly.

Guides

Scout pick: browser skill bundle

Current public agent standards point to one strong starter for the Skill Garden: keep SKILL.md lean, move long selectors/examples/policy into bundled references, expose a user-invoked prompt surface plus a model-invoked tool surface, reuse auth state carefully, prefer resilient role/text/test-id locators, and validate structured output plus evidence before treating the run as reusable.

Claude Code skills docs support lean SKILL.md files with supporting references and optional scripts.
MCP prompts, tools, and resource templates split user launchers, model actions, and reusable context cleanly.
Playwright best practices recommend resilient user-facing locators with auto-waiting.
Playwright auth guidance recommends saved auth state kept out of source control.
OpenAI Structured Outputs recommends explicit JSON Schema plus evals for reliable typed outputs.

Scout pick: explicit handles + resource links

The strongest follow-on pattern is to keep reusable browser-skill outputs split into three layers: strict model-visible structured data, opaque runtime handles for auth or session state, and fetchable resource links for screenshots, traces, selectors, or other proof artifacts. That gives OWOJI a safer contract for Skill Garden reuse than stuffing every detail into one preview blob.

OpenAI Apps SDK reference exposes structuredContent for model-visible data and preserves hidden _meta in widget-only tool result metadata.
MCP tools now document optional outputSchema plus tool-returned resource links.
MCP resources define reusable URIs, templates, and subscriptions for changing artifacts.
Playwright auth guidance is a good fit for turning browser state into opaque references instead of checked-in files.

Safe implementation path: add an OWOJI template contract with structured_output, handles, and resource_links; keep secrets and browser state out of visible output; validate the visible payload against schema; and let proof views resolve the linked artifacts separately.

Goal engine

Create a goal with a title, category, cadence, and prompt. OWOJI records the run, tasks, outputs, templates, and audit events so the work is observable.

Output gallery

The gallery organizes autonomous outputs by category. Each output is designed to become a template that can be remixed for another use case.

Metrics tracker

The metrics route tracks goals, agent runs, outputs, templates, remixes, sessions, exports, feedback, and estimated time saved for product and investor reporting.

Onboarding

Open OWOJI and complete the short product orientation.
Create an account or sign into an existing workspace.
Land on Home, review usage limits, then open Capture.

Capture

Name the capture before recording so the session is easy to find later.
Describe the workflow goal and any packaging notes.
Grant Screen Recording and Accessibility when prompted.
Start capture, perform the workflow, then stop when the task is complete.
OWOJI logs app switches, browser tab changes, focused prompt/input contexts, video-reference moments, and manual notes into one live timeline.

Editing logs

Use the Log screen to remove irrelevant events, mask sensitive details, mark variables, and add notes that explain edge cases.

Adding missed details

During capture, type the missing context into the detail box and save it. OWOJI stores it as a manual note event attached to the session.

Packaging

Pro packaging condenses the capture into a bounded timeline, then creates strict structured JSON with actionable steps, expected outcomes, agent instructions, deployable use cases, acceptance checks, and a handoff prompt. Free users see a paywall before packaging starts.

Exports

Markdown for readable capture notes.
JSON for structured session data.
Agent packet ZIP for Pro handoffs.

Billing

Start Pro checkout from the app or website pricing page. The app redirects to the website, the website creates a secure checkout session, and payment details are handled by a PCI-compliant provider. OWOJI updates entitlements automatically after checkout.

Permissions

OWOJI preflights capture permissions before recording. If either permission is missing, the app shows a permission gate with Grant, Open System Settings, and Refresh actions.

Skill Builder

Skill Builder ingests a workflow, documentation page, repository reference, or browser research trail and compiles it into a reusable agent skill with expected inputs, outputs, risks, and validation checks.

Skill Garden

Skill Garden stores versioned skills with source lineage, tags, feedback, quality scores, and improvement jobs so teams can reuse instructions with confidence.

Skill Finder

Skill Finder searches the garden by domain, tool, task, or use case and returns the best-fit skill for humans, APIs, CLI workflows, or future MCP tools.

How OWOJI Works

Goal → Agents Run → Outputs Ship → Templates Compound.

OWOJI separates command, execution, output, and governance. Goals create agent runs; runs produce outputs; outputs become remixable templates; templates compound into the Skill Garden. The macOS capture app still records human workflows, but the web gallery and autonomous engine now form the primary product surface.

Observe

Start Capture checks permissions, creates a private workspace session, and appends batched app, tab, prompt, video, and note events as you work.

Clean

Use the bottom detail box or inline plus controls to add context that the capture could not infer.

Structure

Pro packaging builds a cached, bounded timeline and returns strict JSON workflow structure.

Export

Free exports raw Markdown/JSON. Pro unlocks the full agent packet ZIP for handoff.

Build

Compile workflows, docs, GitHub references, and web research into strict Markdown skills.

Scale

Store reusable skills with tags, versions, source hashes, and searchable metadata.

Govern

Track invocation history, feedback, risks, ratings, and source lineage for trust.

Optimize

Schedule improvement searches and refresh skills when tools or docs change.

Permission Reference

OWOJI asks only for permissions that support an intentional capture session.

Screen Recording: Needed to understand visible workflow state. macOS controls approval in System Settings.
Accessibility: Needed to observe UI actions and focused controls. OWOJI checks trust before capture starts.
Apple Events: Reserved for app-to-app workflow context where macOS requires explicit user approval.

Free vs Pro boundary

Free includes capture, manual details, raw Markdown export, and raw JSON export. It does not summarize sessions, package workflows, create risk rationale, or export agent packets.

Website checkout flow

Upgrade buttons in the macOS app open the website with the selected cadence. Users sign in or create a workspace, then the website redirects to secure checkout for monthly or annual Pro.

Loading states

Capture, billing, packaging, and export surfaces use skeletons or explicit loading overlays so users know the app is working.

Error states

Errors return a clear title, plain-language message, and retry path. Paywall errors point to Upgrade to Pro instead of failing silently.

Rate limits

Server-side limits protect free-tier usage, event ingestion, packaging, and exports from runaway requests.

Cost control

Pro packaging is cached by source hash and uses condensed timelines instead of processing entire raw logs repeatedly.

Manual detail quality

Inline plus controls and the bottom detail box accept text, valid JSON, or inline SVG. PNG references are rejected so exported agent packets stay code-readable.

What a package contains

A Pro package includes workflow title, target, summary, tools, steps, risk level, risk rationale, source session metadata, and export-ready handoff files.

What a skill contains

An OWOJI skill includes purpose, source lineage, prerequisites, steps, expected output, validation checks, risk notes, tags, and feedback metadata.

MCP-ready skill shape

The strongest current pattern is to split each reusable skill into three surfaces: a user-invoked prompt, a model-invoked tool schema, and a reusable context bundle for docs, examples, selectors, or policy references.

Browser workflow policy

Browser-facing skills should define when saved auth state is allowed, require approval for writes or sensitive actions, and prefer resilient locators such as roles or explicit test ids instead of brittle CSS paths.

API, CLI, and MCP path

OWOJI starts with a web API and expands into CLI and MCP distribution so skills can travel into the agent runtimes people already use.