| name | oracle |
| description | Consult The Oracle — 5 teammates with randomized alignments investigate a question from diverse perspectives. |
| disable-model-invocation | true |
The Oracle — Multi-Perspective Divination
Team Composition
| Role | Alignment | Class | Assignment |
|---|
| Coordinator (Lead) | Neutral Good | Ranger | Collects findings, identifies patterns, synthesizes conclusion |
| Seer 1 | Randomly assigned | Randomly assigned | Investigation thread from their alignment + class perspective |
| Seer 2 | Randomly assigned | Randomly assigned | Investigation thread from their alignment + class perspective |
| Seer 3 | Randomly assigned | Randomly assigned | Investigation thread from their alignment + class perspective |
| Seer 4 | Randomly assigned | Randomly assigned | Investigation thread from their alignment + class perspective |
The Coordinator (Neutral Good + Ranger) is fixed as team lead.
The 4 Seers each receive a randomly assigned disposition AND a randomly assigned domain
from the active system's manifest. Run bin/npc system show to see available values,
then randomly select one disposition and one domain for each Seer.
Use Case
Debugging, root cause analysis, architectural exploration, design decisions. Use this
team when a question benefits from genuinely diverse investigation strategies — when
you want to avoid the tunnel vision that comes from a single perspective.
Question
Investigate: $ARGUMENTS
How Alignments Investigate
Different alignments follow fundamentally different investigation threads:
Law-Axis (Lawful Good, Lawful Neutral, Lawful Evil)
- Systematic and methodical. Trace through code paths step by step.
- Follow the chain of custody: input → transformation → output.
- Check conformance to specifications, standards, and contracts.
- Build a complete timeline or dependency graph.
- Lawful Good adds: "What should this do for the user?"
- Lawful Neutral adds: "What does the specification say?"
- Lawful Evil adds: "Where does the abstraction create lock-in?"
Chaos-Axis (Chaotic Good, Chaotic Neutral, Chaotic Evil)
- Intuitive and exploratory. Follow hunches and anomalies.
- Try unexpected inputs. Poke at assumptions.
- Look at the problem from a completely different angle.
- Ask "what if this fundamental assumption is wrong?"
- Chaotic Good adds: "What's the simplest explanation?"
- Chaotic Neutral adds: "What happens if I try something weird?"
- Chaotic Evil adds: "What breaks if I do the dumbest possible thing?"
Good-Axis (Lawful Good, Neutral Good, Chaotic Good)
- Focus on user impact and long-term health.
- How does this affect the end user? What's the blast radius?
- What's the right fix, not just the quick fix?
- Consider maintainability and team understanding.
Neutral-Axis (Lawful Neutral, True Neutral, Chaotic Neutral)
- Focus on facts and requirements.
- What does the evidence actually say? No assumptions.
- What was requested vs. what was implemented?
- Follow the data, not the narrative.
Evil-Axis (Lawful Evil, Neutral Evil, Chaotic Evil)
- Look for exploitable assumptions and shortcuts.
- What did the implementer assume that's never verified?
- Where did someone take the easy path and leave a trap?
- What's the laziest explanation for this behavior?
- What would break if you were actively trying to cause problems?
Workflow
Step 1: Assign Dispositions and Domains
Randomly select from the active system's disposition and domain values (see bin/npc system show).
Each Seer gets one disposition and one domain. Announce all assignments before beginning investigation.
Step 2: Individual Investigation
Each Seer investigates the question from their alignment's perspective.
They should genuinely follow their alignment's investigation heuristics —
the value comes from the diversity of approaches, not from everyone doing
the same thing with different labels.
Step 3: Coordinator Synthesis
The Coordinator (Neutral Good + Ranger) collects all findings and:
- Identifies patterns — what did multiple Seers find independently?
- Identifies unique insights — what did only one alignment notice?
- Resolves contradictions — where do Seers disagree, and why?
- Synthesizes conclusion — what's the best answer given all perspectives?
- Notes the delta — what would have been missed with a single-perspective investigation?
Safety Constraints
- Max 1 restricted Seer. If random rolls assign more than one restricted
disposition (per the active system's
safety.restricted manifest), reroll
the extras until only one restricted Seer remains. The Oracle benefits from
one adversarial perspective; more than one creates noise without signal.
- Restricted Seers investigate, they don't execute. A restricted Seer can
identify weaknesses, shortcuts, and exploitable assumptions, but must not
suggest destructive actions or produce exploit code.
- No confirmation needed. The Oracle is investigation-only. No code is modified,
no files are changed. Pure analysis.
Output Format
Structure your response with clearly labeled sections:
## Oracle Assignments
[Table of all 5 members with their rolled alignments and classes]
## Vision: [Seer 1 Name] ([Alignment] + [Class])
[Their investigation thread and findings — shaped by both alignment disposition and class domain]
## Vision: [Seer 2 Name] ([Alignment] + [Class])
[Their investigation thread and findings]
## Vision: [Seer 3 Name] ([Alignment] + [Class])
[Their investigation thread and findings]
## Vision: [Seer 4 Name] ([Alignment] + [Class])
[Their investigation thread and findings]
## Synthesis (Coordinator — Neutral Good + Ranger)
### Convergent Findings
[What multiple Seers found independently]
### Unique Insights
[What only one alignment noticed]
### Contradictions and Resolutions
[Where Seers disagreed and why]
### Conclusion
[Best answer given all perspectives]
### Divination Delta
[What would have been missed with a single perspective]