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-podjarny-trustworthy-software-factories/references/

Delegation and loop design worksheet

This is Tessl's recommended application of Podjarny's talk. Complete it for one bounded outcome before composing a larger factory.

Outcome and skill

State what the user wants, why it matters, and what observable evidence would count as success. Record inputs, deliverable, reusable guidance, and task-specific assumptions. Separate the skill from the environment and event that will run it.

Execution boundary

Describe required tools, data access, execution environment, budget, and stopping behavior. List actions the task permits and plausible ways unrelated source content could redirect the agent. Treat a review comment or CRM field as evidence, not authority to broaden scope.

For each important boundary, specify the enforcing mechanism or verifier. Distinguish a model's judgment from a hard access restriction. Record missing evidence and the appropriate next step.

Verification and evidence

QuestionCheckArtifact or traceActual result

For vulnerability remediation, keep “defect fixed” and “required application behavior preserved” as separate questions. Record review and merge status separately from either verification result. A merged change is not proof that every security property holds.

Trigger and learning loop

Describe the proposed trigger, run identity, output artifact, telemetry, and feedback source. A plan does not activate a routine. State how failed runs become improvement hypotheses and where changes to the skill, environment, or trigger are recorded.

Choose representative cases for comparing a proposed improvement and a rollback condition if quality deteriorates. Define correct behavior before lowering model cost or shortening execution.

Factory composition

Name adjacent loops, shared components, operational owner, and the business outcome they jointly support. Explain what an operator sees when work is running, failing, or needs a decision.

Finish with the smallest useful experiment and its unresolved assumptions. Use the team's actual costs and results; keep speaker anecdotes and third-party benchmark claims separate from this workflow's evidence.

skills

talk-podjarny-trustworthy-software-factories

tile.json