Timeless Harness

What the company appears to be: Timeless Harness is a self-reported tool for evaluating reusable Agent workflows using an evidence-led governor that compares native baseline performance against workflow treatments. It is described as a single-person project built for the OpenAI 2026 hackathon.

What changed: The author reports that their first design was retired after empirical testing showed it added unnecessary process without improving outcomes. The current version introduces a "Governor" that makes comparisons between native and workflow conditions inspectable, with a focus on lifecycle decisions (keep, narrow, revise, retire, unresolved).

The single most important open question: Does the described Governor actually improve decision-making about reusable Agent workflows in practice, or is it merely a framework for self-evaluation that has not yet been validated through real-world use?

• The description states Timeless Harness consists of two parts:
  • A small Skill that gives an Agent a lifecycle policy: keep, narrow, revise, retire, or unresolved.
  • A Python Governor that makes native-versus-workflow comparisons inspectable.

• The Governor is described as performing these functions:
  • Freezing task, Skill, rubric, starting workspace, evaluator, and cost rule before outputs exist;
  • Running a native baseline and workflow treatment with the same model settings;
  • Keeping executions in separate random temporary roots;
  • Hashing captures, telemetry, final workspaces, file contents, and file modes;
  • Detecting later changes;
  • Rejecting missing telemetry, non-finite cost rules, sensitive workspace files, and unsupported sandbox access;
  • Preparing a condition-label-blind packet containing answers, artifact manifests, and optional evaluator results;
  • Combining reviewer quality with tokens and wall time;
  • Returning a clearly labeled provisional lifecycle candidate.

• The product is presented as not wrapping ordinary work but governing Skills, instructions, memory rules, evaluators, and repair methods.

• Not evidenced: What the actual output or interface looks like beyond the described technical process. No mention of UI, dashboard, API, or user-facing elements.

• The description states Timeless Harness began as a broad method for improving every Agent task but evolved after self-evaluation revealed that an older wrapper design was less effective than native execution.

• It claims to have "retired its first design" based on empirical evidence showing the old wrapper added unnecessary process without improving performance.

• The current version is positioned as a tool that makes reusable workflows "earn the right to stay", rather than promising to make every Agent smarter.

• The author explicitly states: "A workflow should not survive because it sounds wise, took effort to create, or makes an Agent look busy."

• The product does not claim to be a general-purpose Agent improvement tool but instead focuses on governance and lifecycle management of reusable workflows.

• Inferred: The positioning evolved from a broad solution to a narrow evaluation framework focused on validating workflow utility through empirical testing.

• Not evidenced: No stated target customer or ideal customer profile (ICP).

• The description implies the tool is intended for developers or researchers working with coding Agents, particularly those who create reusable workflows.

• The author notes that the tool was built for a hackathon context, suggesting early-stage experimentation rather than a defined market segment.

• Inferred: Potential users could be teams or individuals building and iterating on Agent-based systems where workflow governance is important.

• Not evidenced: No pricing model, monetization strategy, or business model described.

• The description indicates this is a hackathon submission with no revenue or customer data provided.

• The tool appears to be open-source (the repository includes evidence and documentation), but there's no indication of commercial licensing or paid access.

• Built with: git, github, gpt-5.6-sol, hyperframes, markdown, openai-codex, openai-tts, python.

• The Governor uses only the Python standard library and makes no model calls or requires credentials.

• Codex was used as the main engineering environment for implementation, adversarial review, evidence analysis, security review, and product critique.

• GPT-5.6 Sol with ultra reasoning was used through Codex for various tasks including implementation, review, and critique.

• The author reports using separate fresh-context review passes to find defects such as:
  • NaN cost-policy bypass
  • Mutable run evidence
  • Incomplete workspace hashing
  • Weak condition separation
  • Misleading synthetic Demo

• These defects were repaired, covered by tests, or removed.

• Not evidenced: Deployment infrastructure, scalability, integration capabilities, or delivery mechanisms beyond the described development process.

• Not evidenced: No customer data, revenue, usage metrics, or adoption indicators.

• The project is described as a single-person hackathon submission with no external validation or traction signals.

• The repository includes evidence that caused Timeless to retire its first design:
  • 4 fixed tasks;
  • 12 paired trials;
  • 24 outputs with opaque IDs;
  • 72 rubric score rows;
  • 36 recorded pair preferences from three grader result sets;
  • Grader notes, revealed mapping, and SHA-256 checksums.

• The author states that anyone can run one zero-credit command to verify 44 files and recompute the result.

• The product has not yet demonstrated real-world effectiveness beyond its own internal testing.

• Not evidenced: No stated competitors or competitive landscape.

• The description does not mention similar tools, platforms, or methodologies in the space of Agent workflow governance or evaluation.

• The project is positioned as a novel approach to evaluating reusable workflows, but no comparison with existing solutions is made.

• Risk: The product is described as a single-person hackathon submission with no external validation or traction. This raises questions about scalability and long-term viability.

• Risk: The author explicitly states that the evidence supports retiring the tested old wrapper, but does not prove that the new narrow Governor improves unseen tasks — this claim remains unresolved.

• Red flag: The tool is described as having undergone adversarial review and defect repair, suggesting a lack of initial maturity or robustness in its early versions.

• Red flag: The author notes that hashes are not signatures and temporary roots are not operating-system isolation — indicating potential limitations in the evaluation framework's reliability.

• Inferred risk: Without real-world use cases or performance data, it is unclear whether this tool will be effective beyond its own internal testing.

1. What specific workflows or use cases does Timeless Harness aim to govern? How are these defined in practice?
2. Can you provide more details on how the lifecycle decisions (keep, narrow, revise, retire, unresolved) are made and applied in real-world scenarios?
3. Has the current Governor been tested in any production-like environment beyond the internal evaluation?
4. What is the intended path for adoption or integration into existing Agent-based systems?
5. How does Timeless Harness handle edge cases or failures during execution that might not be captured in the current evaluation?
6. Are there plans to expand beyond the current scope of native vs workflow comparisons, or to add more sophisticated evaluation criteria?