Claude-skill-registry task-orchestration

Execute repo work one task at a time using a strict plan → execute → iterate loop tracked in .copilot-todo.yaml.

install

source · Clone the upstream repo

git clone https://github.com/majiayu000/claude-skill-registry

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/task-orchestration" ~/.claude/skills/majiayu000-claude-skill-registry-task-orchestration && rm -rf "$T"

manifest: skills/data/task-orchestration/SKILL.md

Task Orchestration

Use this skill to drive work through a single authoritative YAML todo file:

.copilot-todo.yaml

Source of truth

```
.copilot-todo.yaml
```
— only status tracker.
```
.copilot-todo.md
```
— legacy/migration only (do not update by hand).

Getting Started with Templates

Creating a new plan from template

# Copy the template to your repo root or session folder
cp .github/skills/task-orchestration/template/plan.md.template plan.md

# Edit plan.md with your tasks (use Markdown checkboxes: - [ ] Task name)

# Generate .copilot-todo.yaml from your plan
python3 .github/skills/task-orchestration/scripts/task_orchestrator.py \
  --plan-md plan.md \
  --plan-mode regen \
  init

Creating a .copilot-todo.yaml from scratch

# Copy the template
cp .github/skills/task-orchestration/template/copilot-todo.yaml.template .copilot-todo.yaml

# Edit the tasks section, then initialize
python3 .github/skills/task-orchestration/scripts/task_orchestrator.py init

Template locations:

```
template/plan.md.template
```
— Markdown plan with task checkboxes
```
template/copilot-todo.yaml.template
```
— YAML structure with example task

Loop (plan → execute → iterate)

Repeat until no runnable tasks remain:

Sync / migrate state

Reuse existing
```
.copilot-todo.yaml
```
(default):

python3 .github/skills/task-orchestration/scripts/task_orchestrator.py init

(Optional) Seed/regenerate tasks from
```
plan.md
```
using Markdown task list items (
```
- [ ] ...
```
/
```
- [x] ...
```
).
- If an item starts with an explicit ID prefix like
```
A1. ...
```
  or
```
T-001: ...
```
  , that ID is used.
- Otherwise a stable-ish
```
P-XXXXXXXX
```
  ID is generated from the text.

python3 .github/skills/task-orchestration/scripts/task_orchestrator.py \
  --plan-md plan.md \
  --plan-mode regen \
  init

Start next task

python3 .github/skills/task-orchestration/scripts/task_orchestrator.py next

Plan (short bullets, only for the chosen task)
Execute (minimal diff; run smallest relevant existing build/test)
Update status

python3 .github/skills/task-orchestration/scripts/task_orchestrator.py update <ID> completed --note "..."
# or
python3 .github/skills/task-orchestration/scripts/task_orchestrator.py update <ID> blocked --note "..."

5b) Commit (required when using

ralph_loop.py --require-commit

)

git add -A
git commit -m "<ID>: <short summary>"

Iterate / expand If you discover missing work, add follow-up tasks (don’t silently expand scope):

python3 .github/skills/task-orchestration/scripts/task_orchestrator.py add "Title" --deps <ID1,ID2> --priority Medium --goal "..."

Ralph-style multi-session loop (Copilot CLI)

To run one fresh Copilot run per task (uses

copilot -p

, consuming Copilot requests accordingly):

python3 .github/skills/task-orchestration/scripts/ralph_loop.py \
  --non-interactive \
  --max-steps 10

# or: derive tasks from plan.md
python3 .github/skills/task-orchestration/scripts/ralph_loop.py \
  --non-interactive \
  --plan-md plan.md \
  --plan-mode regen \
  --max-steps 10

Notes:

This does not bypass quotas; it just splits work into multiple sessions.
By default
```
ralph_loop.py
```
enforces one commit per finished task (use
```
--no-require-commit
```
to disable).
Keep
```
--max-steps
```
small to avoid burning requests if a task gets stuck.