From idea-to-code
Tracks task completion in markdown plan files using plan-file-management subcommands. Mandates test verification of observables and evidence before marking tasks done. Activates on *-plan.md files.
npx claudepluginhub humansintheloop-dev/humansintheloop-dev-workflow-and-tools --plugin idea-to-codeThis skill uses the workspace's default tool permissions.
When working from a plan file (markdown file with task checkboxes), follow these guidelines to ensure proper tracking.
Executes tasks from PLAN.md sequentially with human oversight, handling task splitting, clarifying questions, tests, and learning persistence. Useful for deliberate progress on planned work.
Implements tasks from verified plan.md sequentially: writes code, verifies with TDD where applicable, updates plan progress, commits at checkpoints. For structured project execution.
Orchestrates implementation from latest plan.md file: creates tasks, executes sequentially with status updates and per-task semantic commits, ends with codex-review.
Share bugs, ideas, or general feedback.
When working from a plan file (markdown file with task checkboxes), follow these guidelines to ensure proper tracking.
The plan file is the source of truth for task completion status.
Use the plan-file-management skill's subcommands to read and update plan files:
plan-file-management → get-next-taskplan-file-management → mark-task-completeSee the plan-file-management skill for invocation details.
Do NOT rely on internal tracking (like TodoWrite) as a substitute for updating the actual plan file.
THIS IS THE MOST IMPORTANT SECTION OF THIS SKILL.
A task is NOT complete until the Observable outcome is actually observed and the Evidence is actually collected.
If a task has these fields:
Then tests must verify the Evidence - not the agent manually running commands and checking output.
This applies to all test types:
Example task:
- [ ] **Task 1.3: POST /orders endpoint creates order and returns 201**
- Observable: Endpoint accepts order JSON, persists to database, returns 201 with order ID
- Evidence: Integration test POSTs valid order, asserts 201 response, asserts order exists in database
The test must contain assertions that verify the Evidence:
@Test
void createOrder_returnsCreatedWithOrderId() {
var response = restTemplate.postForEntity("/orders", validOrderRequest, OrderResponse.class);
assertThat(response.getStatusCode()).isEqualTo(HttpStatus.CREATED);
assertThat(response.getBody().getId()).isNotNull();
// Verify persisted to database
var saved = orderRepository.findById(response.getBody().getId());
assertThat(saved).isPresent();
assertThat(saved.get().getCustomerId()).isEqualTo(validOrderRequest.getCustomerId());
}
WRONG: Agent writes the endpoint code, assumes it works because the code looks right RIGHT: Tests contain assertions for the Evidence criteria; tests pass = Evidence verified
Tests are the source of truth for verification, not the agent's judgment.
When creating migration or upgrade plans for existing codebases:
Before implementing a task from the plan, review its implementation details for technical flaws:
For each task in the plan:
plan-file-management → get-next-taskplan-file-management → mark-task-completeWhen implementing a documented plan:
BAD:
Complete task 1.1
Update internal TodoWrite
Complete task 1.2
Update internal TodoWrite
... (plan file never updated)
GOOD:
Complete task 1.1
Use plan-file-management → mark-task-complete for thread 1, task 1
Complete task 1.2
Use plan-file-management → mark-task-complete for thread 1, task 2
Apply this guideline when:
*-plan.md, *-tasks.md)- [ ] task checkboxesWhen committing completed tasks:
Before running git commit, always:
git status from the repository root (not a subdirectory)git add path/to/plan.mdWhen committing from a subdirectory:
git add /path/to/plan.md or cd /repo/root && git add .