Update

4 files changed, 14 insertions, 7 deletions

diff --git a/prompts/skills/project-taskwarrior/1-create-task.md b/prompts/skills/project-taskwarrior/1-create-task.md
index 9db6093..259247c 100644
--- a/prompts/skills/project-taskwarrior/1-create-task.md
+++ b/prompts/skills/project-taskwarrior/1-create-task.md
@@ -6,6 +6,7 @@ Use with `00-context.md`. Project name and global rules apply.
 
 - **Every task MUST have at least one tag** for sub-project/feature/area (e.g. `+integrationtests`, `+flamegraph`, `+bpf`, `+cli`, `+refactor`, `+bugfix`).
 - **When an agent creates a task, always add the tag `+agent`** so agent-created tasks can be identified.
+- **Include references to all context required** to work on the task. So that work can be done with a fresh context, every task must list or link everything needed: relevant files, docs, specs, other tasks, or project guidelines (e.g. paths, doc links, `AGENTS.md`, `README` sections). Put these in the task description or in an initial annotation so that an agent starting with no prior conversation has everything they need in the task itself.
 
 ## Add a task
 
@@ -25,3 +26,4 @@ Multiple dependencies: `depends:<id1>,<id2>`.
 
 - Pick or create a meaningful tag for the sub-project or feature.
 - Add dependencies when one task must complete before another can start.
+- When creating a task, add references to all required context (files, docs, specs) so the task is self-contained for fresh-context work.
diff --git a/prompts/skills/project-taskwarrior/2-start-task.md b/prompts/skills/project-taskwarrior/2-start-task.md
index 4e27fa0..3f38bc1 100644
--- a/prompts/skills/project-taskwarrior/2-start-task.md
+++ b/prompts/skills/project-taskwarrior/2-start-task.md
@@ -2,6 +2,10 @@
 
 Use with `00-context.md`. Project name and global rules apply (including one task in progress per project unless the user says otherwise).
 
+## Start each new task with a fresh context
+
+Work on each new task **must begin with a fresh context** — e.g. a new session or a sub-agent with no prior conversation. That way the task is executed with clear focus and no carry-over from other work. The task itself should already contain references to all required context (added when the task was created); read the task description and all annotations to get files, docs, and specs before starting.
+
 ## Mark task as started
 
 When you begin working on a task, **always mark it as started in Taskwarrior** so current work is visible:
@@ -14,5 +18,6 @@ Do this as soon as you start work on the task.
 
 ## Conventions
 
+- Start each new task with a fresh context; rely on the task’s description and annotations for all required context.
 - Run `task <id> start` when you start working on the task, not only when listing or completing.
 - Do not start a second task for the same project while one is already started and not done, unless the user explicitly asks.
diff --git a/prompts/skills/project-taskwarrior/3-complete-task.md b/prompts/skills/project-taskwarrior/3-complete-task.md
index e1dd337..514bdbd 100644
--- a/prompts/skills/project-taskwarrior/3-complete-task.md
+++ b/prompts/skills/project-taskwarrior/3-complete-task.md
@@ -19,7 +19,7 @@ If any of these fail, fix the issues and recheck. Do not mark the task complete
 1. Spawn a **sub-agent** with **fresh context** (no prior conversation).
 2. Sub-agent reviews the diff, code, or deliverables for the task and **reports back** to the main agent (review comments, suggestions, issues).
 3. Main agent **addresses all review comments** from the sub-agent — no exceptions. Fix or respond to every point.
-4. If changes were made, repeat the sub-agent review until the main agent has addressed all comments.
+4. **After addressing comments, spawn another sub-agent** (fresh context again) to **review the code again** and confirm the fixes. If this second review finds further issues, address them and repeat the sub-agent review until the review is satisfied.
 5. Only then:
 
 ```bash
@@ -28,6 +28,6 @@ task <id> done
 
 ## Conventions
 
-- A task is not done until: best practices met, code compiles, all tests pass, **and** all sub-agent review comments have been addressed.
-- Complete with `task <id> done` only after completion criteria and the sub-agent review (and all follow-up fixes) are satisfied.
+- A task is not done until: best practices met, code compiles, all tests pass, all first-round review comments addressed, **and** a second sub-agent review has confirmed the code (or any further issues have been addressed and re-reviewed).
+- Complete with `task <id> done` only after completion criteria, first review, addressing all comments, and the follow-up sub-agent review are satisfied.
 - When completing a task, note which tasks were unblocked (dependents that became ready), if any.
diff --git a/prompts/skills/project-taskwarrior/SKILL.md b/prompts/skills/project-taskwarrior/SKILL.md
index 82ee456..7686efb 100644
--- a/prompts/skills/project-taskwarrior/SKILL.md
+++ b/prompts/skills/project-taskwarrior/SKILL.md
@@ -11,8 +11,8 @@ Taskwarrior tasks are scoped to the current git repository. **Load only the file
 
 | Action | Load |
 |--------|------|
-| **Create task** | `00-context.md` + `1-create-task.md` |
-| **Start task** | `00-context.md` + `2-start-task.md` |
+| **Create task** | `00-context.md` + `1-create-task.md` (include refs to all context required) |
+| **Start task** | `00-context.md` + `2-start-task.md` (start with fresh context; use task refs) |
 | **Complete task** | `00-context.md` + `3-complete-task.md` |
 | **Annotate / update task** | `00-context.md` + `4-annotate-update-task.md` |
 | **Review / overview tasks** | `00-context.md` + `5-review-overview-tasks.md` |
@@ -21,6 +21,6 @@ Always load `00-context.md` first (project name resolution and global rules); th
 
 ## Task lifecycle (overview)
 
-1. Create task → 2. Start task → 3. Annotate as you go → 4. **Completion criteria** (best practices, compilable, all tests pass) → 5. Sub-agent review (fresh context) → 6. Main agent addresses all review comments → 7. Complete task
+1. Create task → 2. Start task → 3. Annotate as you go → 4. **Completion criteria** (best practices, compilable, all tests pass) → 5. Sub-agent review (fresh context) → 6. Main agent addresses all review comments → 7. **Second sub-agent review** (fresh context again) to confirm fixes → 8. Complete task
 
-A task is not done until criteria are met and all sub-agent review comments are addressed. Details are in `3-complete-task.md`.
+A task is not done until criteria are met, all review comments are addressed, and a second sub-agent review has confirmed the code. Details are in `3-complete-task.md`.