- Home
- Skills
- Rsmdt
- The Startup
- Task Delegation
task-delegation_skill
- Shell
168
GitHub Stars
2
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill rsmdt/the-startup --skill task-delegation- reference.md18.8 KB
- SKILL.md25.1 KB
Overview
This skill generates structured agent prompts and coordination plans to delegate complex tasks across specialist agents. It produces FOCUS/EXCLUDE templates, execution strategies (parallel, sequential, or mixed), discovery steps for file-creating agents, and retry/validation rules to prevent collisions. Use it to decompose work into activities, assign clear outputs, and enforce scope boundaries.
How this skill works
The skill inspects the original task and decomposes it into activity-based units with defined expertise, outputs, and dependencies. It applies a parallel-first decision matrix to recommend execution strategy, adds DISCOVERY_FIRST for file-creating agents, and emits standardized agent prompts with FOCUS, EXCLUDE, CONTEXT, OUTPUT, SUCCESS, and TERMINATION sections. It also validates scope, detects potential file path collisions, and generates retry strategies and termination rules.
When to use it
- Breaking down a complex project into discrete activities
- Launching multiple specialist agents in parallel or sequence
- Creating agent prompts with explicit FOCUS/EXCLUDE and context
- Coordinating agents to avoid file path collisions and shared-state conflicts
- Designing retry, validation, and termination logic for agent failures
- Defining execution strategy when dependencies exist
Best practices
- Decompose by activity (what to do), not by role (who does it)
- Default to parallel execution unless dependencies or shared state require sequencing
- Include DISCOVERY_FIRST for any agent that will create or modify files
- Make EXCLUDE explicit to prevent scope creep and duplicate work
- Specify SUCCESS with measurable completion criteria and a maximum retry limit (typically 3)
- Group independent tasks into parallel batches and chain groups sequentially when outputs are required
Example use cases
- Add user authentication: decompose into security analysis, schema design, API endpoints, and UI with mixed execution
- Competitive research: run multiple competitor analyses in parallel, then synthesize results sequentially
- Large feature launch: create parallel tasks for docs, tests, and implementation with discovery-first checks to avoid path collisions
- Code review flow: generate REVIEW_FOCUS prompts with VERIFY and termination rules for approval or revision
- Migration project: plan sequential schema changes where dependencies and shared state require ordered steps
FAQ
Default to parallel; choose sequential if tasks depend on outputs, share writable state, or risk file path collisions.
When should I add DISCOVERY_FIRST?
Add DISCOVERY_FIRST for any agent that will create or modify files or when existing patterns must be discovered before changes.
How many retry attempts are recommended?
Use a conservative limit such as 3 attempts, with explicit blocking conditions listed in TERMINATION to avoid infinite retries.