Sigil: Skill Decomposer

• source file, folder, or fragment,
• user's intended extraction target,
• known source context,
• candidate names,
• source examples,
• target tier guess,
• conversion constraints.

1. Identify source scope and available context.
2. Map sections, behaviors, roles, dependencies, local vocabulary, gates, and outputs.
3. Preserve source boundaries so extraction decisions remain reviewable.

Step 2 - Identify Candidates

4. List candidate reusable capabilities.
5. Score each candidate for coherence, reuse value, coupling risk, tier clarity, and missing context.
6. Reject implementation details that do not generalize into reusable behavior.
7. Avoid splitting one coherent source into too many tiny candidates.

Step 3 - Select Or Gate

8. Select the strongest candidate when evidence is clear.
9. Ask a human decision only when multiple candidates are equally plausible or the extraction target is ambiguous.
10. Block if the source fragment lacks enough context to define a safe boundary.

Step 4 - Define Boundary

11. Define included behavior, excluded behavior, required context, source risks, target tier, and handoff assumptions.
12. Classify source material as keep, rewrite, reject, defer, or needs decision.
13. Preserve important gates from the parent workflow.
14. Remove source-specific names from the target public contract.

Step 5 - Handoff

15. Produce a handoff package for skill-transcriptor when the candidate is coherent.
16. Include target name, tier recommendation, objective, inputs, modes, process outline, outputs, quality risks, anti-patterns, templates needed, and validation notes.
17. If the boundary remains ambiguous, return a block with the minimum decision needed.

• preserve enough source context to justify extraction,
• identify candidate capabilities before selecting one,
• score coherence, reuse value, coupling risk, and tier clarity,
• define included and excluded behavior explicitly,
• reject local implementation details that are not reusable behavior,
• prevent final sigil authoring when boundaries are ambiguous,
• produce a clear handoff to skill-transcriptor when ready,
• avoid replacing sigil-development lifecycle governance.

• extracting from a fragment with no surrounding context,
• turning every source section into a separate sigil,
• hiding boundary uncertainty,
• losing gates from the parent workflow,
• keeping source-specific routing, project names, or artifact names as canonical terms,
• creating the final sigil before the boundary decision is explicit,
• treating implementation details as reusable capabilities.

• source type,
• candidate count,
• selected candidate,
• rejected candidates,
• coupling risks found,
• boundary confidence,
• decision gate needed,
• handoff target,
• validation result.

## Skill Decomposer Result

- Source: <path or description>
- Mode: map | extract | boundary | handoff | validate | observe
- Candidates found: <count>
- Selected candidate: <name or none>
- Target tier: formulae | transmutations | arcana | undecided
- Boundary decision: pass | flag | block
- Coupling risks: <count>
- Handoff to skill-transcriptor: yes | no
- Files planned: <paths or none>
- Validation: pass | flag | block | not run
- Next action: <action>