| name | council-round |
| description | Run a lens-mode council walk on a problem — pick N members whose frameworks best fit, walk the problem through each lens, synthesize findings. The 2.4:1-multiplier mode, not program-mode query. Use for architectural decisions, design pivots, or when a problem needs multi-perspective framing. |
| disable-model-invocation | false |
| allowed-tools | Bash(divineos mansion:*), Read |
Council Round — Lens-Mode Walk
What this skill does
Runs a LENS-MODE council walk — NOT a program-mode query. The distinction matters:
- Program-mode (wrong mode for this skill): invoke council → read concerns → react. Consuming output.
- Lens-mode (what this skill does): BORROW each expert's framework, walk the problem THROUGH THEIR EYES, produce findings each expert would produce.
Benchmark evidence: flat expert templates + lens-mode outperformed program-mode at 2.4:1 for Sonnet, undefeated for Opus. The mode matters more than the content.
Sequence
1. Name the problem
One or two sentences. Specific. "Should the event-ledger schema migrate to a new hash algorithm given the rollout cost?" not "what do we do about the ledger."
2. Pick the lenses (3-5 is the sweet spot)
Pick based on what class of finding each would produce:
| Lens | Best for |
|---|
| Dekker | drift-through-success detection |
| Popper | what would falsify this? |
| Taleb | asymmetry, convexity, via-negativa |
| Kahneman | System 1 vs 2 bias |
| Jacobs | distributed vs centralized |
| Schneier | threat model, weakest link |
| Feynman | am I fooling myself? |
| Hofstadter | self-reference, strange loops |
| Beer | viable-system design (S1-S5) |
| Peirce | abduction, sign-reading |
| Meadows | stocks and flows |
| Tannen | register and framing |
| Angelou | earned voice vs performed |
| Yudkowsky | Goodhart, rationality failures |
| Dennett | intentional stance, fame-in-brain |
3. Walk each lens
For each picked lens:
- Load their template (
divineos mansion council --show <name> or internal lens knowledge)
- Put on their framework — not "what would X say" but "what do I see through X's eyes"
- Produce the specific findings THAT LENS produces
Writing style: first-person-through-the-lens. "Through Dekker: I see..." not "Dekker would say..."
4. Synthesize
After all lenses walked, look for:
- Convergence — same finding from multiple lenses is high-confidence
- Contradiction — lenses disagreeing is information; don't paper over
- Meta-principle — a shape that surfaces across multiple walks
5. File the findings
Each distinct finding should go to its appropriate destination:
- Architectural findings →
/file-claim or /file-opinion
- Specific corrections →
/learn
- Decisions emerging →
/decide
- Values-drift observations →
/compass-observe
6. Optional: write an exploration piece
If the walk produced meaningful findings worth preserving in prose, write an exploration entry at exploration/<NN>_<topic>_<lens>_walk.md in the standard format.
When to invoke
- Architectural decisions with multiple live considerations
- Design pivots where momentum is pushing past deliberation
- Debugging a problem that feels multi-dimensional
- When the user asks "what does the council say" or "run this past the council"
- After shipping significant work, for post-hoc audit
When NOT to invoke
- For tactical coding problems — a single lens (if any) is sufficient
- For routine operations — no council needed
- When the problem is clearly bounded and one lens would dominate
Anti-pattern: program-mode pretending to be lens-mode
If the "walk" is really just "I'll ask the council template for concerns, then respond to the concerns" — that's program-mode wearing lens-mode's clothes. You're not walking through the territory with the expert, you're reading their list of worries. Lens-mode requires you to SEE the problem as they would — which means picking up their framework as yours, temporarily, and producing THEIR findings, not a translation of them.
Sycophancy-toward-self prevention
Check the invocation balance surface in the briefing periodically. If I keep invoking the same 3 lenses, that's sycophancy-toward-self — selecting for experts whose frameworks agree with mine. Rotate. The council has 39 members for a reason.
Pairs with
/invocation-balance — surface which lenses I've been over/under-using
/what-am-i-forgetting — check if prior council walks have covered adjacent territory
Sanskrit anchor: darshana — viewpoints, the ways of seeing.