# Consolidation matrix

Selection record prepared from the [capability inventory](capability-inventory.md) and actual local/upstream comparisons before integration implementation. These are dispositions, not a claim that the selected changes are already delivered. Source anchors and exact file mappings are in [provenance](upstream-provenance.md).

KEEP preserves the local authority. ENHANCE changes selected guidance without replacing that authority. ADOPT adds a missing capability. COMPOSE reuses a technique inside its existing owner. REJECT excludes redundant/conflicting behavior. DEFER records useful work outside this integration.

## Agentic remains canonical

| Existing capability | Disposition | Evidence and concrete boundary |
| --- | --- | --- |
| Universal/language standards and generated configurations | KEEP | `CODING.md`, seven language directories and initializer golden scenarios; upstream preferences cannot replace them |
| `setup` and initialization CLI | KEEP | Existing dry-run/skip/force+backup and seven language contracts remain; only dependency-complete skill packaging may grow |
| `codinary` | ENHANCE | Preserve all [invariants](codinary-invariants.md); add optional approved-design intake/routing and planning reference; direct requests remain supported |
| `code-review` | ENHANCE | Remove uninstalled setup dependency; require tracker only for tracker fetch. Keep two axes and standalone report contract; never substitute it for Codinary runtime QA |
| `tdd` | ENHANCE | Retain red/green, public seams and all companion assets; selected catches/misses note is useful. Honor already authorized seams rather than repeated gates |
| `diagnosing-bugs` | ENHANCE | Retain loop-first six phases and HITL asset; upstream mutation proof improves red evidence without replacing local conventions |
| `codebase-design` | KEEP | Main skill identical upstream; retain local CONTEXT vocabulary and selective design-it-twice use |
| `domain-modeling` | KEEP | Retain CONTEXT/map/ADR contract; no forced rename to upstream GLOSSARY filenames |

## Matt: every skill disposition

Paths below are relative to `skills/` in the pinned Matt source. Behavioral inputs, outputs, side effects and prerequisites are itemized in the inventory.

| Source | Disposition | Canonical owner / rationale |
| --- | --- | --- |
| `engineering/ask-matt` | REJECT | Agentic registry/routing owns discovery; importing this router would expose missing and competing commands |
| `engineering/code-review` | ENHANCE | Existing `code-review`; selected standards discovery/parallel clarity only, no second copy |
| `engineering/codebase-design` | KEEP | Existing identical skill and local references remain |
| `engineering/diagnosing-bugs` | ENHANCE | Existing debugger; mutation verification wording only |
| `engineering/domain-modeling` | KEEP | Existing local CONTEXT convention and ADR assets |
| `engineering/grill-with-docs` | COMPOSE | `grilling` + canonical domain-modeling; document composition or metadata alias, no wrapper engine copy |
| `engineering/implement` | REJECT | Codinary already owns implementation, independent QA and release; upstream auto-commit conflicts |
| `engineering/implement-spec` | COMPOSE | Extract dependency frontier, ownership and self-contained slice handoffs into Codinary reference; reject autonomous merges/resets/PR closure and duplicate execution loop |
| `engineering/improve-codebase-architecture` | DEFER | Useful optional deepening survey; existing codebase-design plus EGM review already cover chosen design, no survey required for ordinary delivery |
| `engineering/pr` | DEFER | Helpful evidence-first presentation; Humanlayer/show-me secondary credits need their own license/provenance review; no PR workflow needed here |
| `engineering/prototype` | DEFER | Useful separate discovery artifact; relaxed tests, branch commits and production folding require an explicit prototype scope and Codinary boundary |
| `engineering/research` | COMPOSE | Primary-source findings and shared context pointers inside discovery; no new mandatory background-router command |
| `engineering/retro` | DEFER | Optional continuous improvement, not a mandatory delivery stage; keep outside current command growth |
| `engineering/setup-matt-pocock-skills` | REJECT | Agentic setup remains canonical; adopted grilling has no tracker/label setup dependency |
| `engineering/tdd` | ENHANCE | Existing `tdd`; one catches/misses clarification, no GLOSSARY migration |
| `engineering/to-spec` | COMPOSE | Capture requirements, scope and tests in existing brief/design artifacts; EGM owns complex specs; reject automatic issue publishing and parallel spec truth |
| `engineering/to-tickets` | COMPOSE | Codinary planning reference: complete demoable slices, real blocking edges and expansion/contraction when necessary; no tracker publishing subsystem |
| `engineering/triage` | DEFER | Issue/PR labels, comments and closure are valuable but externally mutating and require tracker configuration absent from current integration |
| `engineering/wayfinder` | COMPOSE | Distinguish decision dependencies from implementation slices in Codinary planning; reject map/issue/branch automation and alternative orchestrator |
| `engineering/wizard` | DEFER | Human-only operational setup useful later; bundled secret-handling shell library needs isolated review and explicit procedure scope |
| `productivity/grill-me` | COMPOSE | Optional alias to `grilling` in registry/docs only; one interview implementation |
| `productivity/grilling` | ADOPT | One standalone requirements specialist; inspect available facts, resolve decision frontier, honor scope/authorization and capture downstream criteria |
| `productivity/handoff` | REJECT | Codinary already has durable brief, append-only ledger and handoffs; no competing run-memory format |
| `productivity/teach` | DEFER | Educational mission/lesson system outside engineering framework |
| `productivity/to-questionnaire` | DEFER | Useful asynchronous stakeholder elicitation, unnecessary for this integration; no sending authorization implied |
| `productivity/wait-what` | REJECT | Plain-language clarification can be a communication preference; no canonical command needed |
| `productivity/writing-for-agents` | COMPOSE | Use progressive disclosure and explicit completion criteria when maintaining skills; no additional framework command needed |
| `misc/git-guardrails-claude-code` | DEFER | Frozen/harness-specific hook mutation; Agentic authorization rules already prohibit unauthorized destructive Git |
| `misc/migrate-to-shoehorn` | DEFER | Frozen TS-specific test dependency migration unrelated to integration |
| `misc/scaffold-exercises` | REJECT | Frozen course/vendor lint workflow plus unconditional commit, outside scope |
| `misc/setup-pre-commit` | DEFER | Frozen JS tooling defaults cannot replace seven-language existing policies |
| `in-progress/chief-of-staff` | REJECT | Alternative goal orchestration/schedules; competes with Codinary and runtime lifecycle |
| `in-progress/claude-handoff` | REJECT | Harness-specific background process and duplicate memory mechanism |
| `in-progress/loop-me` | DEFER | Personal recurring workflows outside engineering mandate |
| `in-progress/setup-ts-deep-modules` | DEFER | Useful optional dependency discipline but package reshaping/tool installation is a separate migration |
| `in-progress/writing-beats` | REJECT | Article composition does not fill an engineering capability gap |
| `in-progress/writing-fragments` | REJECT | Prose exploration does not fill an engineering capability gap |
| `in-progress/writing-shape` | REJECT | Article shaping does not fill an engineering capability gap |

## EGM: adopt the distinct design lifecycle

| Capability / source | Disposition | Contract and integration condition |
| --- | --- | --- |
| `skills/egm-brainstorm` | ADOPT | Idea → readiness-confirmed multi-lens concepts brief; no production implementation |
| `skills/egm-draft` | ADOPT | Brief/notes → section-reviewed design, alternatives, constraints and implementation plan; human authority preserved |
| `skills/egm-review` | ADOPT | Five distinct sequential stages and standalone presets; fresh reviewers and traceable reports, no automatic design approval |
| `skills/egm-resolve` | ADOPT | Persistent Elephant resolves fresh findings and applies authorized document edits; no source writes or self-approval |
| `skills/egm-review-loop` | ADOPT | Re-resolve semantic scope each round; stable findings, at most five rounds unless user extends; no interaction-gate bypass |
| `skills/egm-bootstrap` | ADOPT | Optional bounded documentation maintenance; preserve existing notes/alternate filename; failed scouts remain incomplete |
| `skills/egm-drift-report` | ADOPT | Optional current source/document comparison; discrepancies produce report only |
| `agents/*.md`, standards, templates, schemas, icons/metadata, required runtime docs | ADOPT | Complete dependency bundle, single authoritative source; per-skill install dependency closure required |
| `.claude-plugin/agents/*.md` | ADOPT | Runtime wrappers only; one registered wrapper/persona per name, canonical personas remain shared |
| Upstream host/plugin identities and marketplace catalogs | REJECT | Single Agentic plugin/registry owns packaging; translate relevant capabilities instead of installing a second plugin |
| `distributions/claude/` runtime duplicate | REJECT | Generated copies are not canonical and would drift |
| Upstream website, fonts, marketing and contribution automation | REJECT | Not runtime dependencies; retain only referenced runtime visual assets/notices |
| Upstream packager and native test programs | COMPOSE | Reuse reviewed safety/test techniques in local tooling where justified; no blind execution, global installs or auto-updates |

## Selected architecture and stop condition

The intended public set is 15 canonical skills: seven existing, `grilling`, and seven EGM capabilities. Optional aliases route to the same implementation and must not create duplicate registrations. Design artifacts feed an explicit identity/version/revision/checksum-bound handoff; only authorized human approval unlocks approved-design delivery. Codinary remains available without that handoff.

EGM design review is separate from source review, runtime QA and post-release verification. No subsystem can reset Codinary's two-fix count or infer release authorization. Routing depends on uncertainty, affected interfaces, security/data risk and reversibility, not line count. Small work bypasses full design review; optional bootstrap/drift/improvement work is never mandatory for every task.

Implementation begins only after this inventory/selection record and the architecture proposal have been reviewed. Preserve public contracts, validate selected-install dependency closure and single registration, run baseline plus focused regression checks, and report native runtime evidence separately from static validation. Deferred capabilities do not block this integration; missing mandatory runtime evidence remains a disclosed limitation or blocker.

## Specialist extension (2026-10-29)

The 15-skill baseline above is extended to 17 by [Caveman and Security Audit](caveman-security-audit.md). Adopt Caveman core presentation only; exclude its infrastructure and alternate skill aliases. Adopt the complete Security Audit standalone directory, including phases, all attack companions, validators, schema and tests. Compose security guidance with scoped engineering and keep full audit records separate. Codinary ownership, budgets, review-only scope, QA and release authorization remain unchanged.