Refine

SkillMedia

Use when the user asks to run or test Arcanum refine, create a refinement seed, refine a vague target through the canonical discovery/design loop, or inspect the refine lifecycle using native skill execution and dispatch-spec validation.

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 Refine skill

What this skill tells your AI

The instructions your AI receives, as published by cyberalchemyai/arcanum in arcana/refine/development/codex-skill-install/SKILL.md and read by ahel’s review.

Refine is the Arcanum sigil for designing an initial seed, validating a dispatch-spec route for the canonical discovery/design refinement loop, indexing stage evidence, and returning a final refined synthesis.

Use this skill when the user asks for refine, /refine, a refine seed, a refinement run, or a test of the refine workflow.

Local Source

In the domainspec-core checkout, the canonical refine contract lives at:

arcanum/arcana/refine/SKILL.md

Read that file first. It is the source of truth for presets, research policy, dispatch route validation, native stage execution, run-manifest contract, evidence-index requirements, runtime handoff, and quality bar.

Also read only the supporting file needed for the request:

  • arcanum/arcana/refine/REFINEMENT-LOOP.md for loop phases, budgets, stage configuration, repair rules, and research bounds.
  • arcanum/arcana/refine/templates/ when materializing a refinement run folder.
  • arcanum/arcana/refine/examples/ when the user wants an example or comparison.
  • arcanum/arcana/refine/development/run-validation-fixtures.sh when validating fixture-level behavior.

Execution Surface

Prefer the installed native skill package and the canonical source contract:

  • active package: refine
  • canonical source: arcanum/arcana/refine/SKILL.md
  • deterministic helper, when explicitly used: legacy compatibility or handoff preparation only

If the deterministic helper is unavailable, follow arcanum/arcana/refine/SKILL.md directly and record the missing helper as a compatibility-surface gap. Do not require .codex/commands/ or slash-command resolution for normal Refine execution.

Stage Dispatch

Refine should first write REFINE-DISPATCH.json and validate it through formulae/dispatch-spec/dispatch.schema.yml. Native runtime-backed stages run only after the route shape, technique references, gates, handoffs, and observability grouping are valid.

Refine then uses the parent native skill/subagent surface for stage work. When a durable adapter handoff is useful, record that handoff as compatibility or receipt preparation; do not count command-interface execution as native stage proof.

Record each owning capability, capability handle, receipt kind, receipt artifact or fields, requested mode/config, artifact path, observer status, verdict, and blocked reason in the manifest/index.

Runtime Boundary

The durable Arcanum runtime remains the execution target. When dispatch validation, capability resolution, or runtime execution cannot run, Refine records BLOCK with exact missing fields rather than silently switching to Task Session, Sigil Development, or local fallback.

Output

Report:

  • target,
  • preset and research mode,
  • whether native skill or deterministic capability resolution worked,
  • dispatch route validation status,
  • runtime handoff status,
  • run manifest path when created,
  • evidence index path when created,
  • dispatch route path when created,
  • runtime handoff path when created,
  • validation command and result when run,
  • final synthesis or remaining blocker.

Signals

GitHub stars
25
Forks
3
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
refine-cyberalchemyai
Source
github.com/cyberalchemyai/arcanum