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-davis-agent-tool-policies/references/

Generated-tool governance design

This worksheet applies Davis's principles; its fields are recommendations rather than a transcript-provided schema.

Task and effect boundary

Authorized task: Describe what the user wants and which effects are already authorized.

Generated program: State why code mode is useful, what runtime features it needs and which external tools it orchestrates.

Execution choice: Choose an OS sandbox with gateway or a constrained interpreter with harness dispatch. Explain the relevant trade-off rather than asserting one is universally safer.

Authority map

For each tool or external effect, record the caller identity, allowed scope, policy enforcement point and trusted parameters. Show where credentials live and whether generated code can read them.

Describe every route to the filesystem, network or another service. Mark direct paths that bypass the intended dispatcher or gateway as findings requiring verification, not assumed exploits.

History-aware decision

Trusted prior state: Identify the form, approved request or other authoritative source defining the task. Include how legitimate changes are recorded.

Request comparison: Explain how each tool call is checked against that state. For example, a travel booking's beneficiary should match the authorized request unless a valid update changes it.

Outcomes: Specify allow, deny or the existing review path for undecidable cases. Preserve prior user authorization; avoid adding an approval on every harmless step.

Evidence and verification

Record generated program identity, tool request, decision rationale, policy version, relevant state and actual outcome. Minimize sensitive content where possible.

Propose tests for an unauthorized beneficiary, a direct network bypass, missing history and permission expansion. State what result would demonstrate each boundary, and mark all unexecuted tests as planned.

Open questions: List runtime support, policy integration and deployment facts requiring current documentation. End with the selected design and its remaining trust assumptions; do not describe the worksheet as a security certification.

skills

talk-davis-agent-tool-policies

tile.json