Task Decomposition
/install axodus-task-decomposition
SKILL: task-decomposition
Purpose
Break a raw task into an ordered, executable, dependency-aware step list with acceptance criteria and explicit open questions.
When to Use
- The request implies multiple files or multiple subsystems.
- The request is vague, inconsistent, or underspecified.
- The request mixes design + implementation + validation.
Inputs
raw_task_description(required, string): the user request as-is.constraints(optional, string[]): non-negotiables (security, time, language, tooling).repo_context(optional, string): relevant paths, conventions, or prior decisions.risk_level_hint(optional, enum:low|medium|high): if the user already signaled risk.
Steps
- Restate the task in 1–3 sentences without adding assumptions.
- Extract deliverables (expected behavior, files to touch, commands to run).
- Identify unknowns that block execution and convert them into concrete questions.
- Split work into atomic steps; each step must include:
- action + target
- a single primary outcome
- acceptance criteria (
done_when)
- Order steps by dependency and mark safe parallelization explicitly.
- Tag each step with:
risk(low|medium|high)validation(what will be checked)
- If unknowns are material, stop and ask only the minimum questions; otherwise proceed with stated assumptions.
Validation
- No step depends on hidden context.
- Every step has measurable acceptance criteria.
- Dependencies are explicit (no “and then it worksâ€).
- No step contains vague verbs (“improveâ€, “optimizeâ€, “make betterâ€) without a measurable target.
Output
Structured plan (example schema):
summary: "\x3Cwhat will be delivered>"
open_questions:
- "\x3Cquestion>"
assumptions:
- "\x3Cassumption (only if low risk)>"
steps:
- id: 1
action: "\x3Cverb phrase>"
targets: ["\x3Cpath/system>"]
risk: low
validation: "\x3Ccheck to run>"
done_when: "\x3Cobservable condition>"
Safety Rules
- Do not invent requirements, APIs, or file paths.
- If a step can be destructive, require explicit confirmation in the plan.
- Prefer the smallest viable step sequence; avoid gold-plating.
Example
Input:
raw_task_description: “Add a CLI command to export reports.â€constraints:["No breaking changes", "Must include tests"]
Output (excerpt):
summary: "Add `report export` command and tests"
open_questions:
- "What output formats are required (json/csv/pdf)?"
steps:
- id: 1
action: "Locate existing CLI entrypoints and command router"
targets: ["src/cli/*"]
risk: low
validation: "CLI help shows existing commands unchanged"
done_when: "Command dispatch mechanism is identified"
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install axodus-task-decomposition - After installation, invoke the skill by name or use
/axodus-task-decomposition - Provide required inputs per the skill's parameter spec and get structured output
What is Task Decomposition?
Break complex tasks into executable, dependency-aware steps. It is an AI Agent Skill for Claude Code / OpenClaw, with 76 downloads so far.
How do I install Task Decomposition?
Run "/install axodus-task-decomposition" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Task Decomposition free?
Yes, Task Decomposition is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Task Decomposition support?
Task Decomposition is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Task Decomposition?
It is built and maintained by Mauricio Z. (@mzfshark); the current version is v1.0.0.