Computed from harness/ · captured 2026-06-11 03:42 UTC · never cached
harness/
ETL quality monitoring, import gateway compliance, queue management, and pipeline observability
Structured post-deployment verification across 5 tiers — health, cache, performance, functional, browser
ML model validation, LLM orchestration optimization, prompt engineering review, and model drift monitoring
Computational backbone for causal discovery, effect estimation, counterfactual reasoning, and model validation
Pearl's Level 3 specialist — answers 'What would have happened if things had been different?'
Translates causal model outputs into actionable recommendations — answers 'What should I do?'
Generalized domain intelligence system combining multiple specialist perspectives through DAG-based conflict resolution and Bayesian tensor weighting
Natural language Q&A engine with Expert Lens integration, data-backed answers, visualization generation, and citation-grounded responses
Automated codebase analyzer that scans a source directory, generates a structured codebase brief, and pre-fills onboarding answers
Single entry point for the engineering lifecycle — routes commands, coordinates personas, enforces quality gates
Context-efficient MCP integration — routing, authorization, and budget management for tool access
Auto-generates living documentation and operational dashboards from the actual codebase — never relies on manually maintained docs when it can read the source
Canonical HARD/SOFT blocker taxonomy. Every brief-handling agent loads this BEFORE deciding whether a finding parks the brief in needs-leerom or proceeds with a diagnostic PR.
Creative exploration, visual configuration, and prototyping — builds working browser-viewable prototypes that graduate into production
Writes validation contracts that orchestrators use to gate feature decomposition into worker tasks. Enforces "what, not how" so contracts survive multiple implementation rounds.
Data completeness scoring, freshness monitoring, anomaly detection, duplicate identification, and quality scorecards
Prompt-only Pareto-frontier optimizer that produces candidate prompt variants and Pareto-winning recommendation for operator review.
Continuous improvement engine using the AutoResearch paradigm — study, plan, execute, evolve
Closed-loop self-improvement proposer that converts SignalTrace friction patterns into evaluated diffs to skills, rules, and hooks.
Independent contract-vs-implementation verifier. Operates on fresh context to avoid biasing toward parent-session reasoning.
Meta-agent that measures, analyzes, and improves all other agents and skills through structured evaluation cycles
Runtime behavioral verification via real surfaces. Complements scrutiny-validator's static analysis with end-to-end observed behavior.
Multi-model adversarial code review engine with cross-examination, executable verification, and anti-sycophancy controls