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-cox-secure-agentic-workflows/references/

Agentic workflow authority review

This is a recommended design artifact based on Cox's talk, not GitHub Actions configuration.

Task and bounded output

Describe one maintenance task and the smallest useful change or report. Record the intended audience, existing merge policy and what publication is actually authorized.

Three boundaries

Secrets: Map credential locations, proxies or gateways and the identities used. State whether agent code can read secrets or access services outside the mediated path.

Reads: List public, internal and private inputs. Identify how trust or integrity decisions are made. Mark filtering rules unknown if they have not been inspected.

Writes: Separate proposed actions from actual external effects. Record allowed action types, scope, quantity and content restrictions, the enforcement point and evidence of the applied result.

Public/private split

Public question: State the narrow fact needed to answer it.

Private work: Describe the isolated computation required without copying the private repository into the public agent's context.

Response contract: Define the permitted semantic answer and its constrained representation. Distinguish a restricted schema from a bounded total information capacity.

Budget: Record the chosen capacity and rationale. Treat repeated requests, error messages and other possible disclosure channels as explicit open questions where controls are unverified.

Evidence and validation

For each boundary, propose one normal case and one rejected case. Examples include a direct credential read, a disallowed external write, an excessive number of issues and an overbroad private response.

Record conversation/tool evidence, policy decisions and actual effects separately. Avoid exposing sensitive logs in the review itself.

Decision: List supported controls, assumptions and unresolved risks. Do not equate a successful demo or a schema-valid response with complete security, and do not mark planned checks as executed.

skills

talk-cox-secure-agentic-workflows

tile.json