diff --git a/docs/architecture/README.md b/docs/architecture/README.md
index 6f179ca11b..3e751510f2 100644
--- a/docs/architecture/README.md
+++ b/docs/architecture/README.md
@@ -139,7 +139,7 @@ The plan parser (`shared/egg_contracts/plan_parser.py`) extracts tasks and PR me
 2. **YAML front matter** (legacy): A `---`-delimited YAML block at the document start. Supported for backwards compatibility.
 3. **Markdown regex** (fallback): Parses `[TASK-X-Y]` patterns from markdown. Fragile and may miss tasks if LLM output format drifts.
 
-The parser also extracts optional PR metadata (title and description) from the `pr:` field in the YAML data. If provided, this metadata is used when creating the pull request during the implement phase.
+The parser also extracts optional PR metadata (`title`, `description`, `test_plan`, `manual_steps`, `context_title`, `context_description`) from the `pr:` field in the YAML data. If provided, this metadata is used when creating the pull request during the implement phase; the `context_title` and `context_description` fields feed the dedicated context PR rather than the implementation PR.
 
 **Role-aware task assignment**: Tasks can include an optional `role` field (`coder`, `tester`, or `documenter`) that assigns the task to a specific execution agent. The parser propagates this field through `ParsedTask` into the contract `Task` model. During the implement phase, `_build_role_context()` filters tasks by role so each agent only sees its assigned work. See [Agent Roles Reference](../reference/agent-roles.md#role-aware-task-assignment) for details.