Skip to content
Adaptive Skills Docs
Esc
↑↓navigate↵open⌘Jpreview
On this page

Execution Patterns for Skills

Reference documentation for Execution Patterns for Skills in Adaptive Skills.

Execution patterns define which shape of execution a task runs in. They are selected by AletheIA before any skill executes (see the AletheIA repo, docs/contracts/execution-pattern-selection.md). Skills do not select patterns — they declare compatibility with them.

Three axes, not one

Axis Question Declared in
Execution pattern which shape — topology (single agent? fan-out? loop? tournament?) AletheIA pattern selection; skills declare compatibility here
Execution mode how deep — basic / extended / high-risk / multi-agent execution-modes.md
Autonomy level how much authority — observe / advise / act_with_approval / autonomous_within_bounds harness-requirements-for-skills.md

Choosing a pattern says nothing about depth or authority. A skill running as one unit of a fan_out_and_synthesize stage still runs at its own mode and within its own autonomy envelope.

The ten patterns

manual_prompt, single_agent, classify_and_act, fan_out_and_synthesize, adversarial_verification, generate_and_filter, tournament_compare, loop_until_done, scheduled_stateful_loop, human_led_workflow.

The library — what each is and when to use it — is owned by the AletheIA repo (docs/concepts/execution-pattern-library.md). This repo does not restate it; it declares how skills relate to it.

What a skill declares

A skill ships a skill_execution_patterns declaration (template):

  • compatible_patterns — patterns the skill can run in, each with conditions and the controls that pattern requires (e.g. loop_until_done only with an objective gate and max_iterations);
  • incompatible_patterns — patterns the skill must not run in, with a rationale (e.g. feature judgment has no objective stop condition, so loop_until_done is out);
  • required_evidence_by_pattern — what evidence each pattern demands from this skill;
  • escalation_triggers — conditions that route the run back to AletheIA (objective_gate_missing, touches_sensitive_area, judgment_required, recurring_failure, comprehension_debt_risk);
  • audit_requirements — what the run must log (skill id, pattern, controls, evidence refs).

Skills declare, they do not enforce

A skill does not decide alone that it can run in any topology, and it does not execute policy. The declaration is consumed upstream (AletheIA selects the pattern and verifies compatibility) and enforced downstream (harness and enforcement apply limits). This mirrors how harness requirements work: declaration here, authority there.

Was this page helpful?