@dreb/coding-agent 2.65.0 → 2.66.1

package/README.md

@@ -131,6 +131,8 @@ See [docs/providers.md](docs/providers.md) for detailed setup instructions, incl
 
 **Custom providers & models:** Add providers via `~/.dreb/agent/models.json` if they speak a supported API (OpenAI, Anthropic, Google), including Bearer-only Anthropic-compatible endpoints. For custom APIs or OAuth, use extensions. See [docs/models.md](docs/models.md) and [docs/custom-provider.md](docs/custom-provider.md).
 
+**Output limits and retries:** `maxTokens` is the ordinary request maximum, including model overrides; explicit bounded internal calls may use less. Responses truncated at the model's configured output limit retry at that same limit. After exhaustion, genuine output-budget failures fail loudly, while possible full-context truncation enters one compact-and-retry recovery. The ChatGPT-backed `openai-codex` protocol rejects client-supplied output limits, so its built-in maximum remains server-controlled and dreb rejects `maxTokens` overrides for that provider.
+
 **Reasoning across model switches:** Exact-model signed, encrypted, or redacted reasoning state is replayed unchanged. Structured reasoning is portable only between models that share a provider and the `openai-completions` API when the destination accepts the source's recognized plain field (`reasoning_content`, `reasoning`, or `reasoning_text`). Other readable reasoning is retained as labelled plaintext in `<reformatted-pre-switch-reasoning>` markers with incompatible protocol metadata removed; opaque redacted or encrypted-only state is omitted for incompatible targets. This outbound conversion does not change session history, so switching back can replay the original state unless it was compacted or pruned. Custom models must also share provider identity and compatible API/signature behavior.
 
 ---
@@ -413,6 +415,8 @@ Set `backgroundAgents.maxConcurrentSubagents` in `/settings`, dashboard Settings
 
 **Per-agent model overrides:** The model used by each agent type can be overridden via the `agentModels.models` setting (a map of agent name → ordered fallback list) without copying or editing the agent definition `.md` files. Configure it in `settings.json` or via `/settings` → **Agent Models**. Resolution order: per-invocation `model` override → `agentModels` setting → agent definition `model` → parent session model. See [docs/agent-models.md](docs/agent-models.md).
 
+**Single model mode:** The `singleModelMode` setting (default `false`) forces every subagent to run on the parent session's model. When enabled, per-invocation `model` overrides, `agentModels` fallback lists, agent-definition models, spawn-time probes, and the dispatch arbiter are all bypassed, and any requested model selection — even one identical to the parent's — is reported to the parent by prepending a `[WARNING: ...]` note to the child's output. If nothing requested a model, no warning appears. If the parent session has no model, the spawn fails loudly. Configure it in the TUI (`/settings`), dashboard Settings, or `settings.json`. See [docs/agent-models.md](docs/agent-models.md#single-model-mode).
+
 **Optional Dispatch Arbiter:** A global-only `subagentArbiter` setting can enable a fully headless, direct model call after the proposal above resolves but before each child process spawns. It uses the validated routing guide, live explicit model scope, and bounded title-setter-style parent activity including useful tool outputs; existing secret scrubbing applies before inference. It may change only the existing agent type, exact scoped provider/model, and supported thinking level. It has no tools, cannot rewrite the task/cwd or agent definitions, and fails closed on configuration, guide, scope, inference, output, or validation errors. It runs once per single/parallel child and after `{previous}` substitution for every chain step. Interactive `/settings` and dashboard Settings both expose enable/disable, exact model selection, thinking, guide path, and loud readiness/validation feedback. Safe typed changed/unchanged/failure records are persisted outside parent model context and relayed to the TUI, JSON/RPC, and dashboard. See [Dispatch Arbiter](docs/agent-models.md#dispatch-arbiter) and [settings](docs/settings.md#dispatch-arbiter).
 
 **Model identity in system prompt:** The parent session's running model is exposed in its own system prompt as `You are running on: provider/id`. This lets the model make self-aware routing decisions (e.g. delegate vision tasks to a multimodal subagent, or use a differently-architected model as a critic). The line updates automatically on mid-session model switches.