Skip to content

■ SIGNALS // RADAR SIGNAL

a header can be a requirement

kiso validates checkpoint headings; Graphify redacts credential-shaped graph labels; omnarad repairs tool-result handling.

■ [!] ON THIS PAGE ▼

self.md radar — 2026-10-22

a header can be a requirement

kiso is now prepared to reject a durable checkpoint because a heading is missing. Graphify is filtering its own graph artifacts, while omnarad has repaired a tool-result path. The scope is narrow: checkpoint contents, rationale labels, output text.

1. kiso makes checkpoint structure a host policy

kiso 0.51.0 adds host-defined contextPolicy.checkpointSections. Every configured heading must appear as an H2; if one is missing, the checkpoint is rejected through the runtime’s existing serialized retry and breaker path.

When the field is absent, 0.50.1 behaviour stays byte-for-byte. The first request and tool schemas are unchanged.

reading: kiso 0.51.0 release notes

2. Graphify redacts credential-shaped rationale text

Graphify 0.9.93 redacts API keys, tokens, Bearer tokens, private-key headers, and credential-bearing URLs when comments or docstrings become rationale labels. The filter runs before the text reaches graph.json, the visualisation, or exports; git SHAs and UUIDs remain intact.

The release also treats a partly read long subagent file as a partial cache miss on the next --update, rather than silently keeping the incomplete result.

reading: Graphify 0.9.93 release notes

3. omnarad adds E2B and fixes tool-result handling

omnarad 0.2.9 adds E2B as a machine-pool provider. It also fixes duplicated and escaped tool results before they reach the model, and resolves machine connection state through the current daemon runtime.

Its macOS binaries require macOS 13 or later.

reading: omnarad 0.2.9 release notes

  • Personal AI OS tools — the control-plane map for personal agents, receipts, memory, and tools
  • Agent memory systems — what agents should remember, what belongs in logs, and how to avoid memory sludge
  • Parallel AI agents — split work across agents without duplicating effort or hallucinating consensus