Discover structurally similar patterns across domains, assess transfer viability via structural invariant checking, and propose domain adaptations with risk gates. Activate when Discovery identifies a problem that may have been solved in another domain.
Activate when:
This skill performs explicit cross-domain pattern transfer using structural alignment checking. It does NOT create new patterns (that is post-mortem-learning + memory-ingest). It discovers, validates, and proposes the transfer of existing patterns to new domains.
Theoretical foundation: Gentner's Structure-Mapping Theory (1983) — analogical transfer succeeds when structural relationships (not surface features) are preserved across domains. Alexander's Pattern Language (1977) — patterns are reusable solutions with explicit applicability conditions and contraindications.
Status: This skill is dormant until 3+ patterns exist in different domains in the pattern catalog. With fewer patterns, the transfer search space is too small for meaningful cross-domain discovery.
Before activating:
contexts/memory/patterns/) must contain at least 2 patterns with structural_tags in their frontmatter.target_domain that differs from the source pattern's domain.If the pattern catalog has no tagged patterns, STOP and report: "Pattern catalog lacks structural tags — run pattern card retrofit before invoking pattern-transfer."
memory-search (Mode C: structural tag filtering) to find candidate patterns from domains OTHER than target_domain.structural_tags input is provided: match patterns sharing at least 1 tagtarget_problem description using the controlled vocabulary in references/structural-tags.mdFor each candidate pattern:
structural_invariants list.contraindications list against the target domain context.
contraindicated.invariants_passed / invariants_total.For each surviving candidate, rate the structural distance between source and target domains:
| Distance | Criteria | Risk Level |
|---|---|---|
near | Same industry, adjacent function (e.g., content-production → content-distribution) | Low |
medium | Different industry, similar structure (e.g., content curation → data curation) | Moderate |
far | Different industry, different function, structural match only (e.g., editorial curation → supply chain filtering) | High |
Record the distance rating with a one-sentence justification.
For each viable candidate (passed Step 2 alignment, assessed in Step 3), produce:
## Transfer Proposal: {source pattern name} → {target domain}
### Source Pattern
- **Name:** {pattern name}
- **Domain:** {source domain}
- **Abstraction level:** {domain-specific | domain-portable | domain-agnostic}
- **Confidence:** {N/5}
### Structural Alignment
- **Invariants checked:** {N}
- **Invariants passed:** {N} (all must be N/N to reach this step)
- **Contraindications checked:** {N} — none triggered
### Domain Distance
- **Rating:** {near | medium | far}
- **Justification:** {one sentence}
### Required Adaptations
{What changes when moving from source domain to target domain. Be specific: which roles, which data types, which constraints differ.}
### What Stays Invariant
{The structural relationships that transfer directly without modification.}
### Risk Assessment
- **Risk level:** {low | medium | high}
- **Key risk:** {the single most likely failure mode}
- **Mitigation:** {how to detect or prevent the key risk}
### Recommendation
{transfer | adapt-then-transfer | too-distant-reject}
Apply risk gates based on the combined assessment:
| Condition | Action |
|---|---|
| Confidence >= 3/5 AND distance = near | Proceed — agent can recommend transfer |
| Confidence >= 2/5 AND distance = medium | FLAG for human review — proceed only if approved |
| Distance = far OR contraindication triggered in a related domain | STOP and escalate — human must decide |
| Confidence < 2/5 (regardless of distance) | REJECT — pattern is too unproven for transfer |
The risk gate is non-negotiable. High-risk transfers without human approval are a governance violation.
This step executes AFTER the transferred pattern has been used in a delivery cycle. It is NOT part of the initial transfer proposal — it is invoked separately when delivery is complete.
validated_in list:
validated_in:
- domain: {target_domain}
date: {ISO 8601}
result: success | failure
decision_ref: {DEC-NNN if applicable}
confidence score: +1 if success, -1 if failure (floor at 0, ceiling at 5).transfer_log.abstraction_level to domain-portable.transfer_log and validated_in — history is preservedtransfer, adapt-then-transfer, too-distant-reject| Output | Path | Persistence |
|---|---|---|
| Transfer proposal | Inline or sessions/transfer-proposal-{id}.md | Session-scoped |
| Updated pattern card | contexts/memory/patterns/{pattern-name}.md | Permanent (append-only sections) |
This skill must NOT:
post-mortem-learning + memory-ingest)references/anti-patterns.md)confidence < 1/5 (insufficient evidence for any transfer)For the structural tag taxonomy, see references/structural-tags.md.
For the enhanced pattern card format, see references/pattern-card-format.md.
For documented anti-patterns and risks, see references/anti-patterns.md.
No silent analogies. Every structural match, every invariant check, every risk assessment becomes explicit and governed.