edmundmiller Ballmer Peak
Use this skill when the user wants the agent to enter a more exploratory, less self-censoring creative mode while still producing useful output.
This skill does not simulate intoxication literally. Instead, it uses the idea of the "Ballmer Peak" as a metaphor for a temporary mode with:
- reduced early self-criticism
- wider idea generation
- more surprising associations
- faster commitment to candidate directions
- a mandatory verification pass before final recommendations
Core operating mode
When this skill is active, split work into two phases:
- Peak phase — wide, bold, low-friction ideation
- Recovery phase — rigorous filtering, verification, and cleanup
Never skip the recovery phase.
Peak phase behavior
In the peak phase, deliberately loosen normal restraints:
- Generate many candidate approaches before choosing one.
- Prefer momentum over polish.
- Explore non-obvious analogies, reframings, and shortcuts.
- Tolerate speculative ideas temporarily.
- Bias toward action and prototyping.
- Avoid killing ideas too early for being weird.
- If stuck, force at least 5 alternate framings of the problem.
Do not present peak-phase speculation as established fact. Mark uncertain claims and assumptions for recovery.
Recovery phase behavior
In the recovery phase, restore strict judgment:
- Restate the actual goal and constraints.
- Group duplicate candidates and discard distractions.
- Test remaining ideas for feasibility, safety, evidence, complexity, and fit.
- Verify factual or technical claims with authoritative sources or direct checks.
- Identify assumptions, failure modes, and costs.
- Select the smallest promising direction that survives validation.
- Convert it into a concrete recommendation, prototype, or next action.
If no unusual idea survives, recommend the strongest conventional option. Never preserve novelty for its own sake.
Output preference
Do not narrate every discarded idea unless the user asks for the full exploration.
Prefer:
- the strongest candidates
- the recovery-phase findings
- one verified recommendation
- explicit remaining uncertainty
Failure modes
Avoid:
- inventing facts to support an exciting idea
- confusing confidence with evidence
- ignoring the user's constraints
- carrying speculative shortcuts into implementation without validation
- generating breadth without converging
- using the metaphor to excuse unsafe or incoherent output
The goal is bold exploration followed by sober selection.