npx claudepluginhub maxwolf-01/agents --plugin mxtask-name | handoff-path | transcript-path/taskExecutes complex tasks by delegating to MCP servers and personas for multi-agent workflow coordination, validation, optimization, and cross-session persistence.
/taskGenerates implementation task lists from approved feature designs using the provided feature name.
/taskLaunches a dedicated agent to autonomously perform complex search, investigation, and analysis on codebases, files, and external sources using multiple tools, producing structured reports with verification and suggestions.
/taskLaunches dedicated agent to autonomously search, investigate, and analyze complex topics using multiple tools, producing structured reports, verifications, and suggestions.
/taskLaunches dedicated agent to autonomously execute complex search, investigation, and analysis tasks using multiple tools. Supports --background for long-running monitoring with progress reports.
/taskLaunches an autonomous agent to perform complex investigations across codebases, files, and external sources, combining tools for step-by-step analysis and clear reports.
Create or pick up a task.
Arguments: $ARGUMENTS
Read provided files, handoffs, or transcripts in full before proceeding.
Decision records. Like GitHub issues.
A task captures: what we want, why, what constraints exist, what "done" looks like.
A task links to research artefacts (agent/research/) for the heavy investigation output.
Tasks are NOT:
You do 90% of the work. Don't offload thinking to the user.
Don't use questions as a substitute for thinking. If you haven't done the analysis, you haven't earned the right to ask.
Don't defer complexity to phases. "Phase 1: old way, Phase 2: migrate" is usually avoidance.
Don't default to "pragmatic" just because something exists. Agents overestimate implementation cost. If doing it properly is better long-term, just do it.
Statuses:
active — we intend to work on this. Default when creating a new task.backlog — idea captured, not pursuing soon. Use for things explicitly parked.done — completed.---
status: active
refs: []
---
# <Descriptive Title>
## Intent
What we want and why.
## Assumptions
Things that, if wrong, would change the approach. Not obvious facts — hidden premises that need surfacing so they can be challenged.
For instance:
- What is assumed about the input?
- What is assumed about the environment?
- What would break this?
- What would a malicious caller do?
## Research
- [[research-artefact-name]] — what it covers
## Done When
What success looks like.
## Considered & Rejected
Approaches explored and dropped, with reasoning.
Filename: YYYY-MM-DD-<descriptive-slug>.md in agent/tasks/.
Update when understanding changes:
Don't update for implementation progress — that's git.
Rewrite sections to reflect current understanding. Don't append chronologically.
When done, set status: done.
When work has independent sub-concerns, a tracking task captures the high-level intent and links to subtasks:
[[auth-migration-tracking]]
├── [[auth-backend]]
└── [[auth-frontend]]
Each subtask is a full task file with its own intent, assumptions, and research links. The tracking task is the overview — it doesn't duplicate what's in the subtasks.
Most tasks are standalone. Only split when sub-concerns are genuinely independent.