Mission
manifest

[MANIFEST_2026] — 6 MISSIONS · 0 OPERATIONAL · EACH ONE OPENS WHEN IT RUNS

1. MISSION_001NIGHTWATCHOvernight agent orchestrator — scheduled multi-agent jobs with traces, cost accounting, evals, and human gates.ENTER →
2. MISSION_002VERDICTAgent eval & cost CLI — golden tasks, calibrated LLM-as-judge, $/task budgets, and a CI regression gate.PENDING
3. MISSION_003REDLINEBounded review agent for regulated content — cited findings, severity, audit log; never approves on its own.PENDING
4. MISSION_004LEDGERAgentic data onboarding — maps a client's messy exports onto a target schema, writes the transforms and the checks that prove them, keeps lineage.PENDING
5. MISSION_005SENTINELPolicy gateway between agents and their tools — least privilege per job, injection defense, data boundaries, and an audit log that would survive a regulator.PENDING
6. MISSION_006PARLEYReal-time voice agent — streaming speech in, tool calls mid-conversation, speech out, under an 800 ms turn budget, with transcript evals.PENDING