Experiments¶
Table of Contents:
Bootstrap Proof Run Result - Concise record of the first maintained bootstrap proof execution, observed issues, and design implications.
Codex Harness End-to-End Test Report - Operational notes and blockers for the Codex harness test flow
Codex Harness Spike Findings - Comprehensive findings from the Codex API harness spike - constraints, learnings, and recommendations
Codex Headless Communication Experiment Plan - Practical next-step experiment plan for Codex headless communication, incorporating research findings while explicitly bracketing premature directions.
Codex Headless Communication Report - Consolidated report on direct headless Codex communication experiments, observed errors, and the currently viable invocation pathway.
OA01.x Cloud Direction Generation Comparison Result - Independent result note for the three cloud-run OA01.x direction-generation attempts, focused on overlap, novelty, subagent evidence, and practical value.
OA01.x Spike Experiment Register - Minimal maintained experiment register for the current OA01.x spike, limited to headless execution patterns, native subagent viability, and low-overhead operator-facing artifacts.
Run SPIKE-10 Agent Coordination Comparison - Operator note for the five-arm SPIKE-10 comparison using direct Codex, native subagents, explicit Codex and Claude assistant CLIs, and the existing tnh-conductor path.
SPIKE-02 Execution Context Comparison - Lightweight result note comparing headless Codex execution contexts for noise, reliability, and practical usability.
SPIKE-03 Native Subagent Smoke Test - Lightweight result note on whether native headless subagent behavior can be observed and captured clearly enough for the OA01.x spike.
SPIKE-04 Narrow Supervisory Comparison - Lightweight comparison note between a direct single-agent pass and the existing supervisory shell run on the same bounded OA01.x design-review task.
SPIKE-05 Minimum Review Artifact Set - Lightweight result note defining the smallest artifact bundle that still made the OA01.x spike runs understandable and reviewable.
SPIKE-06 Native Codex CLI Baseline - Baseline validation of the standalone native Codex CLI before running the prompt-dir orchestration comparison.
SPIKE-07 Codex Home State Dependency - Differential experiment on which HOME-scoped Codex state is required for a successful headless invocation and which state only affects startup noise.
SPIKE-08 Launch Context Environment Contamination - Differential experiment on whether Codex-on-Codex launch noise comes from PTY shape or from inherited execution environment contamination.
SPIKE-09 Prompt Dir Three-Arm Comparison - Comparison of direct Codex, supervisory Codex, and kernel-mediated orchestration on the same bounded tnh-gen --prompt-dir implementation task.
SPIKE-10 Agent Coordination Comparison Plan - Next comparison plan covering native Codex delegation, explicit external workers, Claude worker invocation, and tnh-gen review/process roles.
SPIKE-10 Agent Coordination Comparison Result - Five-arm comparison of direct Codex, native subagents, explicit Codex and Claude assistant CLIs, and the existing tnh-conductor orchestration path on the same bounded implementation task.
SPIKE-10 Conductor Watch Task Brief - Medium bounded implementation task for comparing agent-coordination arms on live operator visibility in tnh-conductor.
Agent Orchestration Spike Testing Sequence - Concise, unambiguous steps to run the Codex CLI spike in a sandbox worktree.
Cloud Run Artifacts - Table of contents for architecture/agent-orchestration/notes/experiments/cloud-run-artifacts
This file auto-generated.