From 0377951a04021e06eb55e3166466fd1d240e8076 Mon Sep 17 00:00:00 2001
From: Tom Boucher <trekkie@nomorestars.com>
Date: Tue, 17 Mar 2026 07:49:14 -0400
Subject: [PATCH] feat: context window size awareness for 1M+ models (#1086)

GSD was designed around 200k context windows. With Opus 4.6 / Sonnet 4.6
shipping 1M context at standard pricing, GSD should adapt its strategies.

Changes:
- Add context_window config option (default: 200000, configurable to 1000000)
- Expose context_window in init output so workflows can adapt behavior
- Update execute-phase.md to note context-size-dependent strategies:
  - 200k: paths-only executor context, 10-15% orchestrator budget
  - 1M+: richer context passing, inline small phases, relaxed /clear
- Context monitor already uses percentage-based thresholds (adaptive)

Users can opt in via config.json:
  { "context_window": 1000000 }

Addresses #1086
---
 get-shit-done/bin/lib/core.cjs           |  2 ++
 get-shit-done/bin/lib/init.cjs           |  1 +
 get-shit-done/workflows/execute-phase.md | 13 ++++++++++---
 3 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/get-shit-done/bin/lib/core.cjs b/get-shit-done/bin/lib/core.cjs
index 5038dc247..d195bfdfd 100644
--- a/get-shit-done/bin/lib/core.cjs
+++ b/get-shit-done/bin/lib/core.cjs
@@ -65,6 +65,7 @@ function loadConfig(cwd) {
     parallelization: true,
     brave_search: false,
     resolve_model_ids: false, // when true, resolve aliases (opus/sonnet/haiku) to full model IDs
+    context_window: 200000, // default 200k; set to 1000000 for Opus/Sonnet 4.6 1M models
   };
 
   try {
@@ -108,6 +109,7 @@ function loadConfig(cwd) {
       parallelization,
       brave_search: get('brave_search') ?? defaults.brave_search,
       resolve_model_ids: get('resolve_model_ids') ?? defaults.resolve_model_ids,
+      context_window: get('context_window') ?? defaults.context_window,
       model_overrides: parsed.model_overrides || null,
     };
   } catch {
diff --git a/get-shit-done/bin/lib/init.cjs b/get-shit-done/bin/lib/init.cjs
index 9c9d35655..3d92bdc6b 100644
--- a/get-shit-done/bin/lib/init.cjs
+++ b/get-shit-done/bin/lib/init.cjs
@@ -31,6 +31,7 @@ function cmdInitExecutePhase(cwd, phase, raw) {
     // Config flags
     commit_docs: config.commit_docs,
     parallelization: config.parallelization,
+    context_window: config.context_window,
     branching_strategy: config.branching_strategy,
     phase_branch_template: config.phase_branch_template,
     milestone_branch_template: config.milestone_branch_template,
diff --git a/get-shit-done/workflows/execute-phase.md b/get-shit-done/workflows/execute-phase.md
index e5c6afe4d..58ddb3a46 100644
--- a/get-shit-done/workflows/execute-phase.md
+++ b/get-shit-done/workflows/execute-phase.md
@@ -191,8 +191,9 @@ Execute each wave in sequence. Within a wave: parallel if `PARALLELIZATION=true`
 
 2. **Spawn executor agents:**
 
-   Pass paths only — executors read files themselves with their fresh 200k context.
-   This keeps orchestrator context lean (~10-15%).
+   Pass paths only — executors read files themselves with their fresh context window.
+   For 200k models, this keeps orchestrator context lean (~10-15%).
+   For 1M+ models (Opus 4.6, Sonnet 4.6), richer context can be passed directly.
 
    ```
    Task(
@@ -652,7 +653,13 @@ Only suggest the commands listed above. Do not invent or hallucinate command nam
 </process>
 
 <context_efficiency>
-Orchestrator: ~10-15% context. Subagents: fresh 200k each. No polling (Task blocks). No context bleed.
+Orchestrator: ~10-15% context for 200k windows, can use more for 1M+ windows.
+Subagents: fresh context each (200k-1M depending on model). No polling (Task blocks). No context bleed.
+
+For 1M+ context models, consider:
+- Passing richer context (code snippets, dependency outputs) directly to executors instead of just file paths
+- Running small phases (≤3 plans, no dependencies) inline without subagent spawning overhead
+- Relaxing /clear recommendations — context rot onset is much further out with 5x window
 </context_efficiency>
 
 <failure_handling>