Long-Horizon Workflows

Custom workflows become a self-driving system when combined with Ava's orchestration, cron scheduling, and cross-agent coordination. This guide explains how to set up autonomous, multi-day operational loops.

The OODA Loop

protoLabs Studio implements a continuous improvement cycle:

OBSERVE          ORIENT              DECIDE              ACT
  |                |                   |                  |
Quinn monitors    Ava triages        Ava creates        Auto-mode
board, PRs, CI    signals, gaps      features with      executes via
protoResearcher   identifies          the right          custom
scans feeds       priorities          workflow           workflows
  |                |                   |                  |
  +--- results feed back into next observation cycle ----+

Each workflow type handles a different class of work. The board is the coordination layer -- workflow A produces a report that informs workflow B.

Workflow Catalog

Operational (recurring)

Workflow	Purpose	Frequency	Model
audit	Read-only code audit	Weekly or on-demand	Sonnet
dependency-health	CVE/outdated package scan	Weekly	Haiku
cost-analysis	Agent spend analysis	Weekly	Haiku
tech-debt-scan	TODO/deprecated/skip inventory	Bi-weekly	Sonnet

Strategic (milestone-driven)

Workflow	Purpose	Trigger	Model
strategic-review	Progress vs goals, gap analysis	End of sprint/milestone	Opus
postmortem	Incident root cause analysis	After blocked features	Opus
research	Deep investigation of a topic	On-demand	Sonnet

Delivery

Workflow	Purpose	Trigger	Model
standard	Full code pipeline	Feature creation	Sonnet/Opus
content	GTM content creation	Content request	Sonnet
changelog-digest	User-facing release notes	After promotion to main	Haiku
swebench	Benchmark evaluation	On-demand	Sonnet

Composing Workflows

One workflow's output becomes another's input through the board:

strategic-review (Opus)
  "We're missing error handling in the webhook layer"
    |
    v
Ava creates 3 features:
  1. research: "Audit webhook error paths" (workflow: audit)
  2. code: "Add retry logic to webhook delivery" (workflow: standard)
  3. code: "Add webhook delivery tracking dashboard" (workflow: standard)
    |
    v
Auto-mode picks them up in dependency order
    |
    v
Quinn runs QA report on completed features
    |
    v
changelog-digest generates release notes

No explicit chaining mechanism is needed. The board is the orchestration bus.

Scheduled Automation with Ava

Ava can create cron jobs that trigger workflows on a schedule. These run during active Ava sessions:

Weekly Health Cycle

Monday 9am:    dependency-health scan
Tuesday 9am:   tech-debt-scan
Wednesday 9am: cost-analysis
Friday 4pm:    strategic-review (week in review)

Ava creates these as features with the appropriate workflow:

create_feature({
  title: "Weekly dependency health scan — week of 2026-03-28",
  workflow: "dependency-health",
  category: "dependencies",
  projectSlug: "system-health"
})

Continuous Monitoring

Quinn handles continuous monitoring via her daily digest (14:00 UTC). protoResearcher handles continuous research feed scanning. Both post to Discord automatically.

For event-driven workflows, Ava responds to signals:

• Feature blocked 3+ times -> creates postmortem feature
• Sprint/milestone completes -> creates strategic-review feature
• Release promoted to main -> creates changelog-digest feature

Self-Driving Pattern

The fully autonomous loop:

1. Quinn observes (daily digest, board monitoring, CI status)
2. Ava orients (reads Quinn's report, checks board state, reads Notes tab)
3. Ava decides (creates features with appropriate workflows based on what she sees)
4. Auto-mode acts (executes features through their workflow pipelines)
5. Results accumulate (trajectory store, fact store, knowledge base, QA memory)
6. Next cycle uses accumulated knowledge (sibling reflections, project knowledge injection)

What Makes It Long-Horizon

Short-horizon: "Fix this bug" (one feature, one workflow, done).

Long-horizon: "Improve our webhook reliability to 99.9%" (strategic-review identifies gaps, research workflows investigate, code workflows implement, audit workflows verify, cost-analysis tracks spend, QA reports confirm improvement, strategic-review checks progress against the 99.9% target).

The system handles this through:

• Projects with milestones -- strategic goals decomposed into phases
• Dependency chains -- features execute in order
• Workflow diversity -- different task types use different pipelines
• Knowledge accumulation -- each completed feature enriches context for the next
• Scheduled reflection -- periodic strategic-review checks if the goal is being met

Setting Up a Long-Horizon Goal

1. Create a project with milestones via MCP or UI
2. Set the strategic direction in Ava's Notes tab
3. Create an initial strategic-review feature to decompose the goal
4. Ava reads the review output and creates implementation features
5. Auto-mode executes them through appropriate workflows
6. Schedule periodic strategic-reviews to check progress

The system evolves its own plan as it learns from execution results.