dsh-codex-context

Configuration

Everything is a config field; all values are optional (schema defaults shown). Add a config: block to the bundle row (codex-context) in your profile's cordis.patch.yml — a patch replaces the row's whole config, so restate every key you care about.

Field	Default	Meaning
auto	true	Automatic windowing (agent/pre-step pressure) and overflow recovery (agent/request-error).
targetActiveTokens	35000	Token budget for the active window; the backwards walk shadows older nodes until the tail fits. When the routed model's context window is smaller than targetActiveTokens / emergencyThresholdRatio, the effective budget is capped at emergencyThresholdRatio × contextWindow so windowing runs pre-emptively instead of waiting for an overflow.
minRetainedNodes	6	Recent surface nodes always retained verbatim.
maxExcerptLength	1000	Character budget per match-centered search excerpt.
searchDefaultLimit	3	search_history result cap when the model omits limit.
searchMaxScanEvents	20000	Safety cap on log events scanned per search call.
emergencyThresholdRatio	0.85	Pressure fraction of the context window above which the engine writes a real model summary (the parachute).
emergencySummarization	true	Master switch for the parachute; false keeps windowing model-free at all times.
summarizationProvider / summarizationModel	''	Fixed route for the emergency summarizer; empty uses the routed request target, then AgentOptions. Configure both as a pair (both empty or both set) or the plugin refuses to load.
emergencyMaxTokens	8192	Output cap for the emergency summarization request.
maxOverflowRetries	1	Extra windowing attempts after a provider-confirmed context overflow.
notesHint	hint text	Line appended under the pinned notes snapshot; empty disables it.
retrievalHint	hint text	Guidance sentence embedded in every window checkpoint.

Invalid values (ratios outside (0, 1], non-positive budgets) fail the plugin at load.

---