CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

ainativedev/aidevcon-2026-nyc

AI Native DevCon 2026 New York — practical agent skills grounded in conference talks

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

application-template.mdskills/talk-sharma-mohapatra-evolving-intent/references/

Evolving-intent interaction contract

Use this worksheet to define behavior before implementing an assistant. It is an adaptation of the talk's examples.

Capability boundary

User outcome: Legacy intent and fallback: Read-only capabilities: Consequential capabilities: Required arguments: Existing authorization: Unsupported requests:

Describe what the tool does independently of the phrases that might invoke it. Name effects on money, external records, or people when relevant.

Current conversation state

Current intended action: Known arguments and their evidence: Ambiguous references: Superseded values: Dependent information now invalid: External action already completed:

For the flight example, a corrected airport may invalidate route, departure time, and price. Keep the old value only as history, not as an equally valid current destination.

Tool and pause contract

ToolInputsResultSide effectExecution boundary

What is shown before a required confirmation: What counts as approval of these arguments: Behavior on denial: Behavior on a material correction: Behavior if confirmation expires: Behavior on tool error or uncertain completion: Recovery if cancellation arrives after commitment:

Use conceptual behavior rather than inventing an interrupt API. Respect existing authorization for routine steps.

Evaluation cases

  1. Equivalent wording reaches the same intended capability.
  2. Two plausible flights produce a relevant clarification.
  3. A corrected airport triggers fresh dependent estimates.
  4. Cancellation before commitment produces no booking.
  5. An uncertain tool result is reconciled before retrying.
  6. Repeated guessing reaches a defined stop condition.

For each case, record expected action, arguments, non-action, state transition, and evidence. Choose project-specific turn and latency budgets.

Review decision

Behavior sufficiently specified: Unresolved tool semantics: Persistence and recovery requirements: Verification needed: Smallest next implementation step:

Do not mark the flow validated until these cases have actually been exercised against the intended implementation.

skills

talk-sharma-mohapatra-evolving-intent

tile.json