Migrate to Tessl

A skill for migrating an existing Claude skill into a published Tessl tile, then running the full optimization cycle to maximize the tile's Quality and Uplift scores on the Tessl registry.

What this skill does

Takes a Claude skill in one of these forms:

• A single .md file (legacy / Anthropic-style flat skill)
• A folder containing SKILL.md (Anthropic-style bundle with references/ and/or scripts/)
• A .skill archive (zipped)
• A folder that's already a Tessl tile (has tile.json) — fast-paths to optimization

...and walks through seven phases to produce a published tile with:

• Quality 100% when registry-reported issues are fixable
• With-context / Impact 100% when the scenario rubrics are controllable
• Uplift multiplier recorded, but interpreted relative to the baseline
• Passed Moderation ✓
• Activation eval showing the skill fires on all generated scenarios

When to use

Trigger phrases that activate the skill:

• "Migrate my skill to tessl"
• "Convert this skill to a tile"
• "Tessl-ize my skill"
• "Publish my SKILL.md to the Tessl registry"
• "Wrap this .md file as a tessl tile"
• "Move my skill to tessl"

Install

tessl install sharaf/migrate-to-tessl

Prerequisites

This skill is an orchestrator — it hands off to four other Tessl skills at specific phases. Before invoking, install:

tessl install tessl-labs/tile-creator
tessl install tessl/skill-optimizer

tessl/skill-optimizer provides:

• tessl__setup-skill-performance — eval pipeline (Phase 6)
• tessl__optimize-skill-performance-and-instructions — optimization orchestrator (Phase 6)
• tessl__compare-skill-model-performance — multi-model variance check (Phase 7)

If a prerequisite is missing when the skill runs, it stops and surfaces the install command — don't try to migrate without the full set.

How it's organized

The seven phases

1. Preflight + inventory — verify CLI/login/skills/project exist, read the source skill, classify its size
2. Scaffold the tile — hand off to tessl__tile-creator
3. Migrate content — copy body into the scaffold; split into references/ if > 5000 tokens
4. Frontmatter + tile.json + README — the first Quality-lift phase (94% → 97%+ comes from here)
5. Initial publish — tessl tile publish v0.1.0
6. Evals + optimization loop — hand off to tessl__setup-skill-performance then tessl__optimize-skill-performance-and-instructions
7. Multi-model variance check + final publish — hand off to tessl__compare-skill-model-performance, then publish v0.2.0

Why an orchestrator?

The official Tessl skills cover specific stages of the workflow well, but they don't compose themselves. Running them ad-hoc you hit the same pitfalls every time:

• Forgetting to create the Tessl project before first eval run
• Leaving the auto-scaffold tile.json summary in place (caps Quality at 94%)
• Using backticked paths for references (lint flags them as orphans)
• Interpreting single-run eval variance as real gaps (wasting edits)
• Chasing model-specific score ceilings as if they were content gaps

This skill encodes the lessons from those mistakes so the migration produces a clean, optimized, published tile.

Eval results (latest known)

Tested across three Claude model tiers on three independently generated scenarios. All three models hit 100% with-context on all three scenarios — the skill content is fully saturated against the rubric.

Per-scenario with-context (Sonnet):

Registry stats: Quality 100%, Uplift ↑2.17x (Sonnet default).

Activation behavior

The skill auto-fires on 2/3 scenarios with Sonnet (the Kubernetes-framed restructuring scenario gets routed by topic rather than by migration-task intent). Content scoring forces activation, so the 100% result holds regardless — the skill works perfectly when invoked.

Built from

The reference sessions that shaped this skill migrated agentic-harness-architect and software-architecture-audit to public Tessl tiles at Quality 100%. The workflow in this skill is the same one used there, with the dead ends pruned and the current lessons baked into the phase references.