diff --git a/.agents/README.md b/.agents/README.md
index 657ab6eac3..371444a89d 100644
--- a/.agents/README.md
+++ b/.agents/README.md
@@ -410,6 +410,10 @@ append-only `Review Log` (UTC ISO 8601, one line per entry):
 
 A plan approved with no changes goes `draft` → `done` directly.
 
+A plan is never staged or committed: the `done` status and its
+`Review Log` are session bookkeeping, not part of the change. Their
+absence from the commit is expected.
+
 ## 7. Connecting `gh` CLI with a token
 
 The `create-issue` and `create-pr` flows need an authenticated `gh`
diff --git a/.agents/skills/implement-plan/SKILL.md b/.agents/skills/implement-plan/SKILL.md
index 9d99e48a6d..e3c4b3d6cd 100644
--- a/.agents/skills/implement-plan/SKILL.md
+++ b/.agents/skills/implement-plan/SKILL.md
@@ -107,12 +107,15 @@ When the implementation is complete, close the plan file first: flip its
 with the issue URL when one exists (standalone mode, e.g.
 `https://github.com/penpot/penpot/issues/NNNN`); in continue/direct
 mode with no issue, just `done` with no invented identifier. Never
-record commit hashes. Then load the **`create-commit`** skill
-and follow its workflow to commit the changes together with the
-closed plan file, so plan and code land in the same commit. Provide a brief summary
-of what was implemented and why, the issue reference (`issue-NNNN`) when
-there is one, and the model name you are running as so the
-`AI-assisted-by` trailer is set correctly.
+record commit hashes.
+
+**The plan file is never staged or committed.** It is session
+bookkeeping, not part of the change: the commit carries the code and
+the tests. Then load the **`create-commit`** skill and follow its
+workflow to commit them. Provide a brief summary of what was
+implemented and why, the issue reference (`issue-NNNN`) when there is
+one, and the model name you are running as so the `AI-assisted-by`
+trailer is set correctly.
 
 ### Step-by-step mode (on request)
 
@@ -124,8 +127,8 @@ per task" — loop one task at a time:
   one commit per task, never two tasks in one commit. Same inputs as
   always: what and why, the issue reference, your model name.
 - After the final task, close the plan file (`Status: done`, one
-  `Review Log` line with the issue URL when one exists) and include
-  it in that last commit.
+  `Review Log` line with the issue URL when one exists). It is never
+  staged: session bookkeeping does not go into the commit.
 - Show the user the result (what changed, files touched, how it was
   verified).
 - WAIT for the user's confirmation before starting the next task.
diff --git a/.serena/memories/workflow/creating-commits.md b/.serena/memories/workflow/creating-commits.md
index a60234819f..1fd50b88de 100644
--- a/.serena/memories/workflow/creating-commits.md
+++ b/.serena/memories/workflow/creating-commits.md
@@ -2,6 +2,12 @@
 
 Commit only on explicit request. Before commit: `git status`; exclude unrelated user changes.
 
+**Never stage a plan file.** A plan is session bookkeeping, not part of the
+change: the commit carries the code and the tests. The `implement-plan` flow
+still closes the plan (`Status: done` plus one `Review Log` line with the issue
+URL); that is not a file to ship, and its absence from the commit is correct,
+not a missing piece.
+
 Do not guess or hallucinate git author information (Name or Email). Never include the
 `--author` flag in git commands unless specifically instructed by the user for a unique
 case; assume the local environment is already configured. Allow git commit to