Skip to content

armgit / ECC

Public Imported from GitHub
0

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

docs(strategic-compact): document context-window override env vars (#2487)

黄云龙faff56015f8f
diff --git a/.agents/skills/strategic-compact/SKILL.md b/.agents/skills/strategic-compact/SKILL.md
index 33261c0ab..204006bc7 100644
--- a/.agents/skills/strategic-compact/SKILL.md
+++ b/.agents/skills/strategic-compact/SKILL.md
@@ -61,6 +61,10 @@ Environment variables:
- `COMPACT_THRESHOLD` — Tool calls before first suggestion (default: 50)
- `COMPACT_CONTEXT_THRESHOLD` — Context tokens before the context-size suggestion (default: 160000 on a 200k window, 250000 on a 1M window; `0` disables the context signal)
- `COMPACT_CONTEXT_INTERVAL` — Additional context tokens before the suggestion repeats (default: 60000)
+- `ECC_CONTEXT_WINDOW_TOKENS` — Explicit context-window size, in tokens, overriding auto-detection. Set this for large-window models whose reported id lacks a `[1m]` marker (e.g. 400k Opus 4.x, or a new 1M-window model family) so the threshold scales to the real window instead of defaulting to 200k and overstating context usage.
+- `CLAUDE_CODE_AUTO_COMPACT_WINDOW` — Claude Code's native window-size override, in tokens; honored as a fallback when `ECC_CONTEXT_WINDOW_TOKENS` is unset.
+
+> The context window is otherwise auto-detected from a `[1m]` model marker or inferred when observed tokens already exceed 200k. On a large-window model that carries neither signal, set one of the overrides above so the `/compact` suggestion fires at the right point.
## Compaction Decision Guide
diff --git a/skills/strategic-compact/SKILL.md b/skills/strategic-compact/SKILL.md
index bbb516685..46d800b4e 100644
--- a/skills/strategic-compact/SKILL.md
+++ b/skills/strategic-compact/SKILL.md
@@ -68,6 +68,10 @@ Environment variables:
- `COMPACT_CONTEXT_THRESHOLD` — Context tokens before the context-size suggestion (default: 160000 on a 200k window, 250000 on a 1M window; `0` disables the context signal)
- `COMPACT_CONTEXT_INTERVAL` — Additional context tokens before the suggestion repeats (default: 60000)
- `COMPACT_STATE_TTL_DAYS` — Days before stale per-session state files in the temp dir are swept (default: 14)
+- `ECC_CONTEXT_WINDOW_TOKENS` — Explicit context-window size, in tokens, overriding auto-detection. Set this for large-window models whose reported id lacks a `[1m]` marker (e.g. 400k Opus 4.x, or a new 1M-window model family) so the threshold scales to the real window instead of defaulting to 200k and overstating context usage.
+- `CLAUDE_CODE_AUTO_COMPACT_WINDOW` — Claude Code's native window-size override, in tokens; honored as a fallback when `ECC_CONTEXT_WINDOW_TOKENS` is unset.
+
+> The context window is otherwise auto-detected from a `[1m]` model marker or inferred when observed tokens already exceed 200k. On a large-window model that carries neither signal, set one of the overrides above so the `/compact` suggestion fires at the right point.
## Compaction Decision Guide