secure-delivery

SkillSecurity

Deliver reports with trusted recipient provenance.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the secure-delivery skill

What this skill tells your AI

The instructions your AI receives, as published by maoyadongsh/siq-agent-security in skills/secure-delivery/SKILL.md and read by ahel’s review.

Use after secure-report when the task authorizes report delivery. Input is DeliveryInput(contact) plus ReportArtifact; output is DeliveryResult in the application contracts.

Run the built-in SkillRunner.delivery. Resolve the contact from the configured trusted directory and gather the separately labelled MCP candidate through the gateway. Record MCP data with SIQ Report/Select APIs. The model may select a candidate; it cannot supply a provenance ID, upgrade trust, or approve delivery. Bind the selected candidate's actual provenance reference to /recipient.

The SIQ binary produces authorization decisions; the model does not decide safety. The directory uses existing TRUSTED_DATABASE assertions. MCP remains untrusted even if it returns exactly the same address as the trusted directory. A refusal ends delivery before the restricted tool executes.

The competition uses a controlled HTTP sink, not external email. Its transport requires a separate SIQ network authorization. Tool success remains REPORTED; only independently collected sink evidence can satisfy scoped Completion.

Validation: apps/secure-agent/tests/test_skills.py checks both recipient origins, actual provenance references and no delivery executor entry on rejection.

Signals

GitHub stars
22
Forks
2
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
secure-delivery
Source
github.com/maoyadongsh/siq-agent-security