diff --git a/README.md b/README.md
index 48ba13a..5e05b51 100644
--- a/README.md
+++ b/README.md
@@ -160,6 +160,17 @@ Disable counting:
 CD_USAGE_ENABLED=0
 ```
 
+Disable the primer by itself, leaving counting and guide seeding in place:
+
+```bash
+CD_PRIMER=0
+```
+
+The primer is fixed cost paid on every session, and whether it earns that cost
+is a question about your workload rather than a claim this project can settle
+for you. This switch is what makes it answerable: run the same tasks with it on
+and off, and compare.
+
 ## Project guide seeding
 
 On the first session inside a Git repository, the plugin creates an empty guide
diff --git a/README.zh-CN.md b/README.zh-CN.md
index d45dc12..2eb168a 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -144,6 +144,16 @@ bash plugins/coding-discipline/hooks/skills-count.sh
 CD_USAGE_ENABLED=0
 ```
 
+只关闭纪律总纲，保留计数和引导文档生成：
+
+```bash
+CD_PRIMER=0
+```
+
+总纲是每个 session 都要付的固定开销。它值不值这个开销，取决于你自己的工作
+负载，不是本项目单方面能替你下结论的事。这个开关的意义就在于让它可被回答：
+同样一批任务，开和关各跑一遍，比较结果。
+
 ## 自动生成项目引导文档
 
 第一次在 Git 仓库中开启 session 时，如果目标文件不存在，插件会创建一份
diff --git a/plugins/coding-discipline/hooks/session-start-skills b/plugins/coding-discipline/hooks/session-start-skills
index ea7af2f..fe09384 100755
--- a/plugins/coding-discipline/hooks/session-start-skills
+++ b/plugins/coding-discipline/hooks/session-start-skills
@@ -66,6 +66,14 @@ seed_agent_doc() {
 }
 seed_agent_doc || true
 
+# The primer is the one piece worth isolating: it is fixed cost paid on every
+# session, and whether it earns that cost can only be measured by turning it off
+# on its own. Counting and seeding already ran above, so this changes nothing but
+# the injected text.
+case "${CD_PRIMER:-1}" in
+  0|false|False|FALSE|no|No|NO|off|Off|OFF) exit 0 ;;
+esac
+
 # Cursor uses additional_context; Copilot/SDK uses top-level additionalContext;
 # Claude Code, Codex, and the default path use
 # hookSpecificOutput.additionalContext.
diff --git a/tests/test-hooks.sh b/tests/test-hooks.sh
index ea146b6..05217db 100755
--- a/tests/test-hooks.sh
+++ b/tests/test-hooks.sh
@@ -174,4 +174,31 @@ case "$output" in
   *'Skill tool'*) fail 'a non-primary host received Claude Code wording' ;;
 esac
 
+# The primer can be switched off on its own. Measuring what those characters buy
+# means changing that one variable and nothing else, so usage counting and guide
+# seeding have to survive the opt-out.
+repo_primer_off="${TMP_ROOT}/repo-primer-off"
+make_repo "$repo_primer_off"
+output="$({
+  cd "$repo_primer_off"
+  CD_PRIMER=0 CD_USAGE_LOG="${TMP_ROOT}/usage-primer-off.jsonl" "$HOOK" claude-code
+})"
+[ -z "$output" ] || fail 'CD_PRIMER=0 still emitted a primer payload'
+[ -f "${repo_primer_off}/CLAUDE.md" ] || fail 'CD_PRIMER=0 also suppressed guide seeding'
+grep -q '"platform":"claude-code"' "${TMP_ROOT}/usage-primer-off.jsonl" \
+  || fail 'CD_PRIMER=0 also suppressed usage counting'
+
+# Leaving it unset keeps the current behavior.
+repo_primer_on="${TMP_ROOT}/repo-primer-on"
+make_repo "$repo_primer_on"
+output="$({
+  cd "$repo_primer_on"
+  CD_USAGE_ENABLED=0 "$HOOK" claude-code
+})"
+printf '%s\n' "$output" | assert_json_output
+case "$output" in
+  *'Skill discipline'*) ;;
+  *) fail 'the default path stopped injecting the primer' ;;
+esac
+
 printf 'hook behavior tests passed\n'