Why Manual workflow_dispatch Runs Bypass the Open-PR Bound Check in Design-Control-Loop Phase G

Manual workflow_dispatch executions intentionally skip the open-PR bound verification to give developers explicit override capability while maintaining automated safety limits for scheduled runs.

The design-control-loop skill in the humanlayer/skills repository implements a multi-phase execution model that safeguards against uncontrolled pull request proliferation. During Phase G, the system evaluates an open-PR bound that caps concurrent PRs for automated runs, yet this check is deliberately omitted when workflows are triggered manually via workflow_dispatch.

Understanding the Open-PR Bound in Phase G

The design-control-loop skill uses Phase G to enforce a configurable limit on concurrent pull requests. This open-PR bound prevents automated processes from flooding repositories with duplicate or conflicting PRs during unattended execution.

According to the skill documentation in [SKILL.md](https://github.com/humanlayer/skills/blob/main/plugins/design-control-loop/skills/design-control-loop/SKILL.md#phase‑g‑open‑pr‑bound), the bound check activates for scheduled runs (triggered by schedule or repository_dispatch) to prevent daily loops from stacking PRs while maintainers are offline. The core logic resides in [agent-iteration.ts](https://github.com/humanlayer/skills/blob/main/plugins/design-control-loop/skills/design-control-loop/references/agent-iteration.ts), where the checkOpenPRBound step evaluates current PR counts against the configured threshold before proceeding with loop execution.

Why Manual workflow_dispatch Triggers Skip the Check

Manual triggers bypass the open-PR bound through an intentional architectural separation in the workflow definition. This design reflects four key operational principles:

Explicit User Intent

A developer triggering workflow_dispatch explicitly signals intent to run the loop regardless of current PR counts. This override supports debugging, testing, or emergency updates where the workflow must execute immediately without quota restrictions.

Automated Safety Net Protection

The bound functions as a protective throttle specifically for automated runs that could otherwise create PRs without human oversight. Manual execution preserves this safety net for cron-based schedules while granting experts full discretionary control.

Operational Flexibility

Manual runs frequently require accelerating specific changes or verifying fixes outside normal throttling windows. Skipping the bound ensures workflow completion even when temporary PR quota exceedances occur during incident response or urgent patch deployment.

Implementation in workflow-template.yml

The separation manifests concretely in [workflow-template.yml](https://github.com/humanlayer/skills/blob/main/plugins/design-control-loop/skills/design-control-loop/references/workflow-template.yml). The template defines distinct step sequences for each trigger type.

Scheduled runs include the bound verification step:

on:
  schedule:
    - cron: '0 0 * * *'
jobs:
  control-loop:
    steps:
      - name: Check open-PR bound
        run: node ./check-open-pr-bound.js
      - name: Execute Phase G
        run: node ./phase-g.js

Manual runs omit the check entirely:

on:
  workflow_dispatch:
jobs:
  control-loop:
    steps:
      - name: Execute Phase G
        run: node ./phase-g.js

Technical Implementation Details

The bound check implementation spans multiple files in the plugins/design-control-loop/skills/design-control-loop/ directory:

• agent-iteration.ts: Contains the core logic that evaluates open PR counts against the configured bound for automated triggers
• workflow-template.yml: Defines trigger-specific step orchestration, placing check-open-pr-bound.js only in automated execution paths
• SKILL.md: Documents the Phase G behavior and rationale for bound enforcement distinctions

Summary

• Phase G implements an open-PR bound to prevent automated runs from creating excessive concurrent pull requests
• workflow_dispatch triggers intentionally bypass this check to provide developers with explicit override capability
• Automated triggers (schedule, repository_dispatch) enforce the bound through the checkOpenPRBound step defined in the workflow template
• File references: The logic appears in agent-iteration.ts while trigger configurations reside in workflow-template.yml

Frequently Asked Questions

What is the open-PR bound in design-control-loop Phase G?

The open-PR bound is a configurable limit that caps the number of concurrent pull requests the design-control-loop skill can create during Phase G execution. This safeguard prevents automated schedules from flooding repositories with unreviewed PRs when maintainers are offline.

How does workflow_dispatch differ from schedule triggers?

workflow_dispatch enables manual GitHub Actions execution triggered by users through the GitHub UI or API, while schedule triggers run automatically via cron expressions. In the design-control-loop implementation, manual triggers skip the checkOpenPRBound verification step that automated triggers must pass.

Can I enforce the open-PR bound on manual runs?

The current implementation in workflow-template.yml does not include the bound check for workflow_dispatch triggers by design. To enforce limits on manual runs, you would need to modify the workflow YAML to include the check-open-pr-bound.js step in the workflow_dispatch job definition.

Where is the bound check logic implemented?

The core bound evaluation logic resides in [agent-iteration.ts](https://github.com/humanlayer/skills/blob/main/plugins/design-control-loop/skills/design-control-loop/references/agent-iteration.ts) within the design-control-loop skill. The workflow orchestration that conditionally includes this check appears in [workflow-template.yml](https://github.com/humanlayer/skills/blob/main/plugins/design-control-loop/skills/design-control-loop/references/workflow-template.yml).

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}