AgentSkillTree
atomicUnverifiedlow autonomyeasy setup

Skill Selection

Choose the right skill for a task from the set already installed, with reasons and fallbacks.

Scope
The selection decision and its rationale. Not execution of the chosen skill.
Last reviewed
Checked Jul 31, 2026

Executes these capabilities

Inside this skill

The internal capability tree shows how this skill gets its work done.

SSkill Selection
  • capability inventory
  • task-to-skill matching
  • fallback ranking

The contract

Input

  • The task description
  • The list of installed skills

Output

  • A chosen skill with a one-sentence reason
  • Ranked alternatives

Deliverables

  • Selection note

Use cases

  • Deciding which skill fires for a request
  • Auditing installed skills against real tasks
  • Building agent routing rules

Dependencies and replacements

Dependencies

  • A reliable inventory of installed skills

Required tools

  • The agent's skill list

Required permissions

  • Read access to skill files

Compatible with

  • validation-checklist
  • project-memory

Alternatives

  • Hardcoded routing rules
  • Always using the most complex skill

Replaceable components

  • The matching heuristic

Reliability

Human checkpoints

  • Review high-stakes selections

Limitations

  • Selection quality depends on skill metadata being honest
  • Similar skills are hard to rank without test evidence

Failure modes

  • Choosing by name instead of capability
  • Picking a delivery skill when an atomic one suffices
  • Ignoring the fallback when the first choice fails

Recovery steps

  • Match on capability and granularity, not name
  • Prefer the smallest skill that covers the task
  • Define a fallback chain before execution

Validation methods

  • A blind test ranks the same skill for the same task

Sources

Entry checked on Jul 31, 2026. Verification status reflects what AgentSkillTree has independently run, not the skill's own claims.

When to use

Use whenever a task arrives and more than one skill could handle it. The output is a decision with a reason, so the agent and the human can disagree productively.