What does this skill do?

The Dynamic Flow Skill Mode designs task-specific local harnesses for one-off processes, extracts reusable skills when a pattern recurs across teams, and establishes observable evaluation gates that validate each step before merging. It transforms chaotic workflows into disciplined systems with durable handoff artifacts and verifiable evidence.

Harness for a one-time task
Generate a local harness with a contract, loop, and evaluation gate for a specific task that does not warrant a shared skill.
Shared Skill Extraction
Promote a repeatable process as a reusable skill when it appears in multiple repositories, teams, or sessions with a stable contract.
Assessment Portals
Depending on the type of work, perform targeted testing, browser smoke testing, fixture transcription, or a claims checklist before approving.
Transfer Items
Record the plan, queue, execution, exit, and handoff so that the task persists across sessions and computers.

Usage examples

🔧 Migration harness
Create a harness to migrate dependencies across three modules with passing tests and a validation gate that blocks breaking changes.
♻️ Skill Promotion
This migration process appears in three repositories following the same pattern; please evaluate whether it should be extracted as a shared skill.
🚪 Evaluation Door
Design a smoke test-style validation gate for a harness that updates a React UI.
📋 Carryover between sessions
Create a handoff artifact that includes status, evidence, and next steps for a workflow spanning two sessions.

Features

Task Decision Tree Classify each task as a one-time inline task, a local harness, or a shared skill based on its frequency and scope.
Explicit Harness Contract Define the objective, inputs, outputs, evaluation, and handoff before writing any code.
Disciplined Harness Loop Detect status, generate a minimum viable artifact, run checks, log the handover, and stop in case of failures.
Evaluation Gates by Type Code-focused test, smoke test for the UI, transcript for agents, checklist for content, dry run for integrations.
Criteria for Promotion to a Skill Level Promote only when there is multi-repo, security sequencing, repeated failures due to skipping gates, or a stable contract.

Frequently asked questions

When the task is one-time but complex: it requires a custom loop, evaluator, crawler, or watcher that does not exist as a shared skill.
When it appears in multiple repositories or teams, errors recur because gates are skipped, or the entry/exit contract is stable and reusable.
The output indicates the blockage, the person responsible, and the necessary action. It is not considered reusable until another colleague can independently rerun the assessment.
No. The skill works with Claude Code and Cowork without any additional configuration, using the agent's native tools.
Dynamic Flow Mode — Reusable Harnesses and Skills with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.