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-braun-context-graphs/references/

Connected-context decision worksheet

This worksheet adapts Braun's talk into a small design and evaluation artifact.

Decision to improve

User task: Delivery stage: Question that local code cannot answer: Current failure or missing evidence: Expected useful outcome:

Choose a specific decision. A general ambition to collect more documents is not enough to evaluate whether the context layer helps.

Sources and relationships

SourceRelevant evidenceLink to work or codeOwnerFreshnessCaller access

Start with a few sources. Distinguish code ownership from organizational decision ownership. Mark uncertain relationships instead of drawing inferred edges as facts.

Retrieval contract

Job-oriented entry question: References needed first: Details fetched only after relevance is established: Evidence included in the final answer: Irrelevant context excluded: Conflict or stale-source behavior: Missing-evidence response: Permission restriction and intended audience:

Specify how the same task behaves when the caller cannot access one source. The correct result may be a partial answer with a clear gap, not a broader search through privileged material.

Five-prompt comparison

  1. Hidden dependency in another repository.
  2. Relevant decision history outside code.
  3. Responsible owner beyond repository ownership.
  4. A question whose answer is absent.
  5. A question with permission-limited evidence.

For each prompt, write expected facts and supporting sources before comparison. Keep current and connected runs comparable. Blind the evaluator to the condition when practical, then inspect material judgments.

Measures: Correctness and coverage: Source support: Irrelevant context: Time and tokens: Unresolved evaluation uncertainty:

Decision and next increment

Observed benefit: Observed failure: Smallest connection or retrieval change: Verification required: Expansion justified by evidence:

Leave results blank until runs exist. Avoid treating an illustrative demo or vendor-reported improvement as a local outcome.

skills

tile.json