Understand AI.Use AI.Build with AI.

Run an after-action review on the event below.

Event: {{describe what happened}}
Goal: {{what we set out to achieve}}
Result: {{what actually happened}}
Timeline: {{rough sequence of events}}

Output:

• What was supposed to happen — restate the plan clearly

• What actually happened — restate the outcome, with key moments

• Why it diverged — root causes, not just symptoms (use "but why?" 3 times)

• What worked well — what to keep doing

• What didn't — what to stop or change

• What we learned — surprises, misconceptions corrected

• Action items — concrete changes for next time, with owners

Rules:

• Blameless tone — we're learning, not assigning fault

• Be specific — generic learnings ("we should communicate better") are useless

• Push back on conventional explanations — challenge assumed causes

• End with: "What would we do differently if we ran this again tomorrow?"

Design a function specification for an AI agent.

What the function should do: {{description}}

Output a JSON Schema-style specification:
{
"name": "verb_noun" (snake_case, narrow),
"description": "What it does + when to call it + return shape (1-3 sentences)",
"input_schema": {
"type": "object",
"properties": {
"field": {
"type": "string|number|boolean|object|array",
"description": "What this field is, valid values, defaults",
"enum": ["..."] (if applicable)
}
},
"required": ["..."]
}
}

Then list:

• Failure modes — what could go wrong; what error messages should be returned

• Idempotency — is the function safe to retry?

• Side effects — what changes in the world when this runs?

• Suggested guardrails — confirmation prompts, rate limits, scope restrictions

Rules:

• Narrow scope — one verb, one noun

• Strong typing — no "any" types

• Self-describing — model should be able to use it correctly first try

Generate unit tests for the function below.

Cover:

• Happy path (with realistic input)

• Boundary cases (empty, single element, maximum size)

• Invalid input (null, wrong type, malformed)

• Edge cases specific to this domain (think hard — what would a senior reviewer flag as untested?)

Use {{test framework}}. Each test gets a descriptive name. Group by category.

Function:
"""
{{paste function}}
"""