/loom

Blueprint Validation Checklist

Purpose: The 14-item gate every operating-layer blueprint must pass before delivery; also the rubric for audit mode. Read when: You are in VERIFY, or auditing an existing layer.

Contents

• Pass criteria
• The 14 items
• Audit-mode usage
• Failure handling

Pass criteria

• All REQUIRED items must pass. Delivery is blocked otherwise.
• RECOMMENDED items pass at ≥ 80%.
• Document any item deliberately skipped, with reason.

The 14 items

Coverage (REQUIRED)

• L1 Every recurring task from SURVEY has a mechanism decision in the coverage matrix (skill / recipe / workflow / hook / scoped rule / none). No task left unmapped.
• L2 Each chosen mechanism is the cheapest that solves the task (no skill where a hook/rule suffices; no workflow where a recipe suffices). Cross-checked against _common/MECHANISM_SELECTION.md (hook/rule/subagent/skill) and recipe-design.md / workflow-design.md (recipe/workflow — Loom-local mechanisms).
• L3 Grounded in real signals — every proposed component traces to an observed task, pain point, or convention in SURVEY, not a generic stack template.

Overlap (REQUIRED)

• L4 Intra-suite overlap < 30% — no two project skills/recipes claim the same primary responsibility.
• L5 Ecosystem overlap handled — tasks owned by a global agent (≥ 30%) are deferred via the routing map, not duplicated as project skills. 30-49% cases were confirmed with the user.
• L6 Routing map is singly-owned — every task domain has exactly one owner; conflicts resolved.

Topology & sizing (REQUIRED)

• L7 Every designed workflow declares a formal topology (hub-spoke / pipeline / hierarchy). No Bag of Agents.
• L8 Every workflow is ≤ 5 phases with verification checkpoints at phase boundaries.
• L9 Agent-count and hub-spoke limits respected (≤ 4 flat / hierarchy beyond; ≤ 7 spokes per hub). Multi-agent use justified (single-agent < 45%) or single-skill chosen instead.

Delegation (REQUIRED)

• L10 Every blueprint component names exactly one owning agent (Sigil / Nexus / Orbit / Latch / Grove) with the payload it receives.
• L11 No authored skill bodies, executed chains, or installed files are present in Loom's output — design and handoffs only.
• L12 Loops deferred to Orbit, state machines to Weave, runtime routing to Nexus — none built inline.

Authoring quality pass-through (RECOMMENDED)

• L13 Each skill spec to Sigil requires a third-person, gerund-form name and a description with negative triggers.
• L14 Each recipe/workflow defines non-overlapping signal keywords / subcommands so Nexus routing is unambiguous.

Audit-mode usage

In audit mode, run the same 14 items against an existing project layer (no generation). Report each failure with evidence and a remediation proposed as a delegation spec — do not apply changes. Typical audit findings:

Failure handling

• Any REQUIRED item fails → blueprint is BLOCKED; fix and re-verify before delivery.
• RECOMMENDED below 80% → fix or document the exception in _STEP_COMPLETE.Output.validation.
• Repeated failure of the same item across two passes → stop iterating and surface the structural cause (usually a missed SURVEY signal or a mechanism mis-pick) rather than retrying.