Multi-Agent Orchestrator 20260731

Overview

This net-new draft converts redacted local multi-agent governance and handoff evidence into a deterministic report. It is advisory and read-only. It does not execute code or agents, access networks, discover credentials, alter files or configuration, control services, schedule jobs, deploy, publish, package, or transact.

When to Use

• Before acting on a multi-agent governance and handoff proposal.
• When a reproducible PASS, REVIEW, BLOCKED, or ERROR report is needed.
• To prepare deterministic fixtures for independent Viper QA.

Do not use it as execution authority, security clearance, release approval, or proof of live health.

Safety Contract

• Reads direct --text or one explicitly named local notes file.
• Refuses hidden or sensitive-looking paths, symlinks, non-files, and inputs over 200,000 characters.
• Uses Python stdlib only and reports mutated_files: [].
• Local PASS reflects stated evidence, not independent verification.

Main Script

scripts/multi_agent_orchestrator.py supports --notes, --text, --self-test, and --json.

Operator Workflow

1. Prepare redacted evidence; complete when no secrets or private personal information remain.

2. Run the analyzer; complete when deterministic JSON is returned without source mutation.

3. Assign every risk and missing-evidence finding; complete when high-risk items have owners.

4. Hand unchanged artifacts to Viper; complete while state remains PENDING_VIPER.

Result Semantics

PASS requires at least nine expected signals and no risk. REVIEW means partial evidence. BLOCKED means two risks or score below 25. ERROR means invalid input.

Test Scenarios

Complete evidence produces PASS; the explicit risk fixture produces BLOCKED; missing input produces missing_notes; guarded and oversized inputs are refused.

Common Pitfalls

1. Treating self-reported evidence as independently verified.

2. Including secret values instead of redacted references.

3. Equating local PASS with execution authority.

4. Omitting ownership, restore, rollback, or acceptance evidence.

Verification Checklist

• [ ] Script compiles with isolated bytecode.
• [ ] Two self-tests are byte-for-byte identical.
• [ ] Self-test returns clean=PASS, risky=BLOCKED, and mutated_files=[].
• [ ] Repository security/static checks report no target findings.
• [ ] Draft remains PENDING_VIPER.

Draft Status

PENDING_VIPER — local Iceman evidence is not an independent downstream verdict.