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-gruenbaum-reverse-graph-engineering/references/

Answer-key-first enterprise evaluation

This is a scenario design template adapted from Gruenbaum's talk. It describes a future test until execution evidence is attached.

Question and oracle

Domain expert's question: Expected answer: Required facts and relationships: Time at which the answer is valid: Deterministic derivation: Correct response when evidence is absent: Information withheld from the agent:

Choose a question that exposes a real reasoning or retrieval failure. Keep the oracle outside the agent-visible data.

Minimal graph

Entities and identifiers: Relationships: Authoritative records: Historical records: Distractors: Cross-system identity mapping:

Build the smallest sufficient company. Add scale only when the evaluation requires it.

Corruption ledger

MutationRecord affectedReason for realismOracle impactExpected agent behavior
Alias or duplicate
Stale or conflicting record
Missing link or migration

If a mutation makes the answer genuinely ambiguous, change the expected response accordingly. Do not punish appropriate uncertainty.

Projection and interface

Documents and tables: Service or protocol views: Pagination and update behavior: Fault cases: Supported operations: Unsupported operations: Permission assumptions:

State the tested subset. A local twin is not the full external service.

Validation and execution

Answer still derivable: Temporal and identity invariants: Distribution or shape checks: Realism review: Scenario and grader versions: Generation seed or recorded choices: Cached artifacts: Held-out scenarios: Trace and score location:

Keep deterministic correctness, perceived realism, and representativeness distinct.

Decision

Failure stage: Evidence from execution: Smallest agent or data-pipeline change: Regression case: Remaining real-world gap:

Do not infer production readiness from a synthetic pass. Record the limits that the next evaluation must address.

skills

talk-gruenbaum-reverse-graph-engineering

tile.json