Deliver an Approved AI Slice

SkillDev tools

Deliver the smallest approved vertical slice of an AI-enabled workflow through real interfaces, users, tests, and operating ownership. Use when workflow, value, data, and mechanism decisions are accepted and the team needs implementation, adoption, validation, and release evidence without expanding scope.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Deliver an Approved AI Slice skill

What this skill tells your AI

The instructions your AI receives, as published by davidahmann/applied-ai-field-guide in .agents/skills/deliver-approved-ai-slice/SKILL.md and read by ahel’s review.

Turn an accepted workflow boundary into one usable, reversible, and supportable vertical slice. Do not reopen mechanism selection or enlarge authority without new evidence and a new decision.

Read first

  1. Read Solution Design and Delivery, the Production Implementation Playbook, and the delivery and adoption plan.
  2. Load the approved workflow charter, value case, data-context manifest, intelligence-selection record, architecture decision, and only the selected solution pattern. Read only the artifacts that apply to the chosen mechanism.
  3. Apply ARC-001, ARC-002, ARC-004, ARC-005, FDE-003, ADP-001, DEL-001, DEL-002, REL-001 through REL-005, and applicable data, tool, security, evaluation, and operations controls.

Workflow

  1. Confirm the approved segment, accepted outcome, verifier, exclusions, maximum effect, authority ceiling, cost budget, safe fallback, adoption surface, owner, baseline acknowledgment, required participant capacity and delegates, and stop conditions.
  2. Translate the boundary into one end-to-end slice across input, decision, human work surface, permitted action or staged artifact, source-of-truth verification, telemetry, support, and rollback.
  3. Build deterministic policy, state transitions, authorization, idempotency, and effect verification outside model generation. Add model or agent behavior only where the selection record requires it.
  4. Use real target interfaces or faithful contract doubles. Preserve source revisions, preparation lineage, release dependencies, and customer-specific configuration.
  5. Build contract, component, trajectory, artifact or outcome, safety, operations, adoption, and recovery tests before expanding effect authority.
  6. Put the slice in the operator's real work surface. Observe exposure, completion, correction, override, abandonment, review load, acceptance, and verified effect separately.
  7. Assemble the compatible release evidence, production-readiness gaps, rollout, rollback, support, receiving-team exercises, and exit conditions. Bind the exact version and prior scope of reused assets and record target revalidation. Merge is not deployment.
  8. Decide iterate, shadow, canary_candidate, constrain, or stop using target-system authority. Do not claim customer acceptance from test completion.

Output contract

Return:

  • the implemented vertical-slice boundary and applicable design packet;
  • tests and evaluation evidence tied to exact source, software, policy, and behavior revisions;
  • operator, adoption, support, recovery, cost, and ownership evidence;
  • unresolved production-readiness gaps, rollout and rollback conditions;
  • one next release or learning decision.

For deterministic, optimization, or classical-ML-only slices, retain equivalent target-software evidence without placeholder agent artifacts. A prototype, demo, merge, or passing local suite is not production approval.

Signals

GitHub stars
105
Forks
22
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
deliver-approved-ai-slice
Source
github.com/davidahmann/applied-ai-field-guide