Openharness Daily: Cleaning Up Agent Memory
The agro repository retired its legacy Claude markdown symlinks in favor of native Agents markdown support, and followed up by formalizing a new memories tier for durable operator context. Both changes, from the same author, streamline how the coding agent stores and reads long-term context.
Duration: PT2M10S
Episode overview
This episode is a short developer briefing from Openharness Daily.
It explains recent repository work in plain language.
- Show: Openharness Daily
- Published: 2026-09-19T12:00:30Z
- Audio duration: PT2M10S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's September 19th, 2026, and this is Openharness Daily.
Today's signal from the agro repository is a single, coherent cleanup effort: rethinking how the coding agent remembers things across sessions.
It starts with PR 1083, which retires five symlinks that used to emulate Claude markdown behavior. The reason: Claude Code's latest release added native support for Agents markdown files directly, reading them at the root and in subdirectories automatically. Keeping the old symlinks around wasn't just redundant —…
That same day, PR 1085 built on the cleared ground. It introduces a formal memories tier — a set of files with clear ownership. One file holds the agent's voice and guardrails, another holds operator identity and standing preferences, and a third is reserved for runtime lessons a session decides are worth keeping.…
Together, these two pull requests tell one story: the team is consolidating on the platform's native memory mechanism instead of maintaining workarounds, and then using that cleaner foundation to add structure for long-term context. Issue 1082 tracked the symlink retirement, and issue 1084 tracked the new memories…
What's next: anyone relying on the…