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

outline.mdskills/talk-davis-agent-tool-policies/references/

Source outline

Read transcript.md for exact wording. Promotion cues 225–235 are omitted.

TimeCuesTopic
00:00–01:451–29Scope, sandboxes, code mode and planned demos
01:48–03:4330–57Untrusted generated code differs from the harness
03:49–05:2658–80Tool orchestration and many calls in generated programs
05:31–08:5481–128Open versus locked-down execution and governance sandbox
08:58–09:49129–140Inherit permissions and policy without expansion
09:54–12:59141–185OS sandbox and separate egress gateway
13:09–15:59188–224Monty constrained interpreter and harness dispatch
16:45–20:28236–289Search and approval-based booking demo
20:36–22:57291–323Skipped demo, forms and execution history
23:00–27:09324–378External policy engine and history-aware denial
27:17–28:35380–398Architecture analogy and no universal winner
28:40–30:18399–425Credentials, egress, inheritance and recording

Short quotations

  • “The generated code isn't the harness” — excerpt, cue 49, 03:10.
  • “Inherit, never exceed.” — excerpt, cue 415, 29:44.
  • “record everything.” — excerpt, cue 417, 29:51.

Limits

No Q&A is captured. The planned crash/recovery demonstration is not shown in this transcript. Two demos are described, not all three initially planned. A controlled form and denial example do not establish general prompt-injection immunity. Monty language support is event-time and partly speculative. The recalled vulnerability comparison at cues 60–62 is an imprecise secondhand recollection, not a usable benchmark.

skills

talk-davis-agent-tool-policies

tile.json