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-poulter-factory-improvement/references/

Factory improvement decision record

This is Tessl's recommended application of Poulter's talk. Complete it before promoting a change to reusable instructions, evaluation, orchestration, or runtime behavior.

Change and hypothesis

Describe the triggering failure and link its actual artifacts or traces. State the reusable unit being changed, intended gain, and plausible downside. Keep the product fix distinct from the proposed factory improvement.

Record inputs, resolved instruction version, model, effort, tools, environment, output contract, and grading method. Explain what can be replayed and what remains variable.

Comparison design

Choose a local comparison with and without the change. State repetitions and why they are sufficient for the decision; the talk's ten is an example. Choose a regression set representing ordinary work and important edge cases beyond the original incident.

Case groupQualityCostLatencyFailure or regressionEvidence

Leave results blank until measured. Include uncertainty and trace inspection where aggregate scores conceal different behavior.

Tradeoff and promotion

Define acceptable gains and tolerable losses before deciding. Record the process version to promote, evidence retained, observation period, and rollback condition. A local success with missing regression evidence remains a provisional result.

Optional runtime decomposition

If the change concerns infrastructure, map model loop, tools, source checkpoints, and code execution separately. Identify isolation, credentials, state ownership, and execution sizing. Compare startup, resource use, and added operational complexity; do not assume that moving tools removes security obligations.

If the change selects tests from a symbol graph, record known coverage gaps and how selective checks will be compared with a broader reference run.

Finish with promote, revise, reject, or gather more evidence. State the strongest evidence, the largest unresolved downside, and the next smallest experiment. Use actual workload measurements instead of copying the talk's reported costs or timings.

skills

talk-poulter-factory-improvement

tile.json