Backstage: The Harness Leaves The Chat Box

This note is product intake for Bitter. The public digest should be useful to any serious builder; this page records what Bitter should learn from the same research.

Bitter Read

The stronger the agent harnesses get, the more important the operator-owned loop becomes. The agent can change. The durable record around the work should compound.

For this window, the key internal interpretation is:

• Agent-side goals are useful, but they are not the project charter.
• Agent-side memory is useful, but it should be reviewable before it becomes durable project memory.
• Visible computers are becoming part of serious agent work, not a novelty.
• Permission and sandbox semantics are product primitives, not security footnotes.
• Accessibility work is distribution, trust, and operator leverage.
• Integrations are volatile; the loop around them needs to be stable.

What Bitter Should Test Next

• Codex /goal as a provider-native goal under a Bitter charter.
• Gemini Auto Memory as a reviewable memory proposal source.
• Agent Zero as a BitterGrid-style visible workcell.
• Paperclip's adapter runtime command spec as a model for run provisioning contracts.
• OpenHands secret/log/sandbox patterns against BitterPass and Grid boundaries.
• OpenClaw setup recovery and channel progress visibility as accessibility benchmarks.
• Pi as a thin, replaceable agent adapter with exact provider and transport records.

Product Implications

Bitter CLI

The CLI should record which agent-side goal, memory patch, skill report, plugin set, and session state shaped a serious run. These are not automatically Bitter truth, but they are relevant run context.

BitterGrid

The workcell should become visible enough that a human can inspect the browser, files, screenshots, runtime state, logs, and artifacts that shaped the work.

BitterPass

Permission profiles should model the real authority surface of each harness: credentials, approval mode, sandbox, network posture, OAuth state, plugins, and known secret-handling caveats.

Factory

Factory should not become the agent. It should own the joined operating view and the run contract: charter, mandate, agent, runtime, authority, cost, evidence, recovery, and next action.

Research Quality Notes

• OpenClaw's high commit volume needs a stronger source contract. The next run should separate product direction from stabilization churn.
• Claude Code was excluded because the current source contract does not define a public commit stream.
• This run was commit-harvest focused. The next version should include release notes, docs, and public announcements where source contracts mark them as primary evidence.
• Selected high-signal commits received diff-level review, but broad commit sampling is not the same as complete review.

Receipts

Findings and signal records for this run are under:

runs/2026-05-07-commit-harvest-2026-04-23_2026-05-07-frontier-v1/