Skip to content
Back to skills

Compile 3d Workflow

ASecurity

Use when the user asks for direction and a compilable 3D workflow from an interview. Not for remote, credential, publish, deploy, or irreversible changes.

  • 52 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 3, 2026
ai-agentsnode

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned September 20, 2026

npx -y skills add OutlineDriven/outline-driven-development --skill compile-3d-workflow --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Compile 3d Workflow?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Compile 3d Workflow
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/outlinedriven-compile-3d-workflow-outline-driven-development/badge)](https://www.skillsdirectory.com/skills/outlinedriven-compile-3d-workflow-outline-driven-development)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: compile-3d-workflow
description: 'Use when the user asks for direction and a compilable 3D workflow from an interview. Not for remote, credential, publish, deploy, or irreversible changes.'
---

# Compile 3D workflow

## Contract

| Field | Bound contract |
|---|---|
| Trigger | The user asks for direction and a compilable 3D workflow from an interview. |
| Authority | Reversible local: writes only the named local 3D workflow artifact file; rollback is deleting that file. No remote mutation. |
| Side effect | Writes one local 3D workflow artifact file. |
| Done | A validated local 3D workflow artifact file exists and passes every structural check. |

## Inputs

- A direction interview with the user, supplying: the problem being solved, what success looks like, binding constraints, and what is explicitly out of scope. All four required; none inferred.
- The output file path for the workflow artifact. Required.

## Artifact schema

The workflow artifact is a YAML file with this structure:

```yaml
problem: <string>
success_criteria: <string>
constraints: [<string>, ...]
out_of_scope: [<string>, ...]
topology:
  nodes:
    - id: <string>
      task: <string>
      depends_on: [<node_id>, ...]
ontology:
  groups:
    - name: <string>
      members: [<node_id>, ...]
feedback_loops:
  - name: <string>
    sensor: <string>
    comparator: <string>
    actuator: <string>
```

Every node id referenced in `depends_on` must exist in `nodes`. Every node id in ontology group `members` must exist in `nodes`. Every feedback loop must name a sensor, comparator, and actuator as non-empty strings.

## Procedure

1. Conduct the direction interview. Ask the user for the problem, success criteria, binding constraints, and explicit out-of-scope. Record the answers verbatim. Done when: all four interview inputs are recorded verbatim, or the missing input is named and the skill stops.
2. Author the workflow artifact from the interview answers. Build the three dimensions:
   - Topology: a DAG of tasks. Each node has an id, a task description, and a depends_on list naming the node ids it waits on. No cycles. No node depends on itself. Every depends_on entry must reference an existing node id.
   - Ontology groups: named concept clusters classifying the work domains. Each group names its member node ids. Every group is non-empty. Every member must exist in the topology.
   - Feedback loops: cybernetic control cycles. Each loop names its sensor (what is measured), comparator (what is expected), and actuator (what action re-routes). All three are non-empty strings.
   Done when: the artifact combines all three dimensions from the interview answers.
3. Validate the artifact against the schema. Check every rule:
   - Every node has a unique id and a non-empty task.
   - Every depends_on entry references an existing node id.
   - No cycle exists in the dependency graph (topological sort succeeds).
   - Every ontology group is non-empty and every member references an existing node id.
   - Every feedback loop names a non-empty sensor, comparator, and actuator.
   Done when: every check passes, or the specific defect is named and the skill stops.
4. Write the validated artifact to the output file path as YAML. State the rollback path: delete the local artifact file. No remote state was touched. Done when: the artifact file is written and the rollback path is stated.

## Failure and recovery

- Interview incomplete: if the user cannot supply the problem, success criteria, constraints, or scope, stop and report which inputs are missing. Do not infer or fabricate direction.
- Schema validation failure: if any node lacks dependencies, any ontology group is empty, any feedback loop is missing its sensor, comparator, or actuator, any depends_on entry references a non-existent node, or a cycle exists in the dependency graph, report the specific defect and stop. Do not emit a partial artifact as complete.
- Partial-result rule: a partially written artifact file is not a compiled workflow. Delete it and report the rollback.

## Output

A validated local 3D workflow artifact file (YAML) containing the DAG topology, ontology groups, and feedback loops, with every structural check passing.

Files in this skill

  • SKILL.md4.2 KB
  • agents/openai.yaml157 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.