Execution Modes

Reference documentation for Execution Modes in Adaptive Skills.

Execution modes define how deep a capability should run.

The rule is: use the shallowest mode that still makes the next step safe, reviewable, and restartable.

Workflow modes

workflow/basic
 └── obvious task, low risk, small proof

workflow/extended
 ├── material ambiguity
 ├── dependencies
 └── multiple validation steps

workflow/high-risk
 ├── irreversible or costly failure
 ├── human decision gate
 ├── traceability
 └── premortem or readiness review

workflow/multi-agent
 ├── multiple specialists
 ├── coordination protocol
 ├── checkpoint and handoff discipline
 └── AletheIA-level governance

Debugging modes

debugging/quick-fix
 └── obvious local failure, low blast radius

debugging/root-cause
 ├── reproducible failure
 ├── boundary check
 └── recurrence guard

debugging/production-incident
 ├── high urgency
 ├── observability-first
 ├── rollback or escalation
 └── human approval boundaries

Governance tradeoffs

Token and resource implications

• Basic mode should avoid extra artifacts unless they change the decision.
• Extended mode may produce a compact plan or trace.
• High-risk mode justifies deeper context, risk mapping, and gates.
• Multi-agent mode must pay for coordination only when role split improves quality.

Anti-overengineering check

Before deepening the mode, ask:

• What risk does the deeper mode reduce?
• What decision becomes more reviewable?
• What evidence would be missing without it?
• Would a smaller checkpoint be enough?

If the answer is unclear, stay shallow.