Tool schemas are the silent context tax: dozens of MCP tools can cost 10k+ tokens per request. The fix is to ship names only and let the model fetch full schemas on demand through a search tool. Both major harnesses converged on this within months (Codex on by default since 0.142).

Models are now trained expecting their reasoning blocks back in subsequent turns. K3's docs warn that dropping thinking history makes generation "highly unstable"; GLM-5 ships an explicit Preserved Thinking mode. Harness and model are co-designed artifacts now.

Every read line carries a content-hash anchor; edits target anchors, batches are atomic, and stale-anchor errors return fresh anchors for immediate retry. The first structural answer to the two classic edit-tool failures: ambiguous matches and files that changed underneath you.

Compaction has always been harness-initiated. D-Mail inverts it: the model decides its recent context is waste, picks a checkpoint, and folds everything since into one note to its past self. Expect others to copy this.

The era of one prompt for all models is over. Codex-tuned models get a third of the words of general GPT-5; OpenCode ships a different dialect per model family; Gemini versions its snippets. Corollary: as behaviour moves into training, prompts shrink toward pure constraints.

Gemini's repo gates edits to the Context Efficiency prompt section behind SWE-bench runs. System prompts have CI now. If you build a harness, version your prompt and regression-test it.

Zhipu treats Claude Code as a commodity interface: Anthropic-compatible API, a model trained to the harness's conventions, a subscription priced per prompt. Kimi and MiniMax run the same play. The moat moved from scaffold to model-fits-scaffold.

Hermes Agent reflects after tasks and writes its own SKILL.md files; /learn converts docs into skills. The skills format (markdown plus frontmatter, lazily loaded) is the cross-vendor convention, and AGENTS.md is now a Linux Foundation standard used by 60k+ repos.

Harnesses tag channel authority explicitly: harness-authored directives, supplementary info, untrusted external content (Gemini untrusted_context, Claude Code system-reminder, Kimi two-tier tags). The model is taught which wrapper outranks which. A security architecture made of XML tags.

dsh applies "everything is a plugin" to the agent itself: loop, model adapter, session log, sandbox, compaction, all swappable, composed per session from preset files. No prompt file, just plugin-owned sections assembled at request time. Claude Code's rules-live-on-tools idea taken to its structural conclusion.

The session is an append-only event log; the model-visible transcript is a projection over it. Compaction is a logged, lock-bracketed transaction (crashes leave detectable orphaned locks), token accounting is replayable via shadow-price events, and 175 snapshot fixtures pin the exact rendered prompt. The harness is regression-tested like a database.

Instead of only borrowing a harness (GLM) or only building one, dsh absorbs the ecosystem: Claude Code and Codex are subagent backends, and their hook formats are supported natively. Your hooks.json works; your existing CLIs become delegation targets. Interop as a wedge for selling V4 tokens.