The bridge never executes commands and never grants permissions. Tool calls, credentials, network access, mutation ordering, and external side effects remain host responsibilities.
Install
npx skillscat add sourabhjainr/auren/agents-skills-agency-runtime-v8 Install via the SkillsCat registry.
Agency Runtime v8 — AI Coding Orchestrator Integration
Use the Agency Runtime as the control plane for substantial coding work.
Contract
The AI coding orchestrator owns provider/tool execution, permissions, repository mutation, concurrency, and side effects. Agency Runtime owns task profiling, specialist planning, evidence, verification metadata, provenance, artifact regression, and release gating.
Required flow
profile -> route -> plan -> execute(host) -> record evidence -> verify -> regression -> release
Before substantial work preserve:GOAL | NON-GOALS | REQUIREMENTS | CONSTRAINTS | PROTECTED BEHAVIOR | BOUNDARIES | ACCEPTANCE | RISKS | ASSUMPTIONS | intent_digest.
Specialist use
Select one primary specialist whenever possible. Add support/reviewer specialists only when independent expertise changes the result or a material risk requires it. Specialist output is work product, not proof of completion.
Evidence
Every material claim must have evidence or be explicitly marked uncertain. Record source, claim, and locator where available. Do not place secrets or sensitive payloads in provenance details.
Verification
Use the normal AUREN order:syntax/static -> focused tests -> integration/system -> regression replay -> security/policy -> final diff review.
The orchestrator must report incomplete checks instead of converting them into success.
Regression
Provide baseline and post-change artifact snapshots for substantive changes when available. Unexpected added, removed, or changed artifacts fail the release decision unless explicitly allowed by task policy.
Provenance
Persist the run trace through portable.agency_provenance.ProvenanceLedger. Verify the chain before treating the run as reusable evidence.
Release
Use portable.ai_coding_agency_bridge.run_coding_task. A run is ready only when the quality receipt, hard gates, findings, artifact regression, and release policy all pass.
Security
The bridge never executes commands and never grants permissions. Tool calls, credentials, network access, mutation ordering, and external side effects remain host responsibilities.