Manor Workspace Blueprint

SkillProductivity

Use when creating, exporting, hand-authoring, installing, upgrading, reverting, reviewing, or debugging any Manor Workspace Blueprint, especially when Workspace behavior differs from its Blueprint or when workflows, tasks, automations, artifacts, approvals, or runtime requirements are involved.

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 Manor Workspace Blueprint skill

What this skill tells your AI

The instructions your AI receives, as published by manor-os/manor-ai in .agents/skills/manor-workspace-blueprint/SKILL.md and read by ahel’s review.

REQUIRED BACKGROUND: Use manor-workspace-architecture first. Map the requested Blueprint change to Workspace runtime nodes and their cross-node contracts before reading or changing export_workspace() or installation logic. Blueprint is the WS-14 portability boundary, not the Workspace architecture itself.

Treat a Blueprint as a portable projection of a real Workspace configuration, not as a second workflow implementation. The authoritative lifecycle is:

Workspace created/configured in Manor
  -> export_workspace()
  -> validated Blueprint payload
  -> install_blueprint()
  -> installed Workspace
  -> export again for semantic round-trip comparison

Non-negotiable rules

  1. Do not hand-edit a Blueprint as the primary way to design Workspace behavior. Configure a real Workspace first, export it, then make only deliberate portable edits.
  2. Never claim install/export parity from validate_payload() alone. That validator checks shape and safety, not runtime equivalence.
  3. Treat Workflow, Proposal-generated Task behavior, and Automation as separate contracts. Include all three in the review and in the round-trip evidence.
  4. Distinguish portable configuration from runtime state. Do not export or promise portability for Task instances, task logs, workflow runs, agent executions, conversations, messages, leases, credentials, measurements, or generated artifacts unless the payload explicitly defines a safe portable representation.
  5. Any non-empty payload section that the installer does not materialize is a release blocker. Either implement its installer/exporter mapping or mark it explicitly excluded and keep it empty.
  6. Natural-language fields (summary, descriptions, prompts, operating-loop copy, acceptance text, and not_included) explain the structured contract; they must not introduce steps, tools, approvals, or deliverables absent from that contract.

Required workflow

  1. Apply manor-workspace-architecture, read its architecture and node test checklist, then read runtime-contract.md and inspect the current exporter, installer, payload validator, and upgrade code.

  2. Define the intended Workspace behavior as a structured inventory: operating model, agents/subscriptions, skills/knowledge, workflows, tasks, automations, goals/stats, governance, channels/sessions, and deliverables.

  3. Create or update the real Workspace configuration. Do not start with a free-form Blueprint JSON unless the task is explicitly a migration.

  4. Export with export_workspace() and run the portable payload validator.

  5. Inspect the exported payload for source-row IDs, secrets, runtime state, and unsupported sections. Fail closed on any of them.

  6. Install the payload into an isolated Workspace in both simulation and live modes when both modes are in scope. Record install todos and materialized IDs; an install with blocking todos is not equivalent.

  7. Export the installed Workspace again. Normalize only generated identifiers, install timestamps, sandbox/provenance fields, and workspace-specific job suffixes. Compare the runtime-projected sections, including Workflow graphs, Automation triggers, artifact contracts, and approval scopes. Verify one-shot install declarations (contract.variables, unresolved channel/session requirements, post_install_checks, and expected_baseline) from install preview/todos and check results; they are not reconstructable from the installed Workspace.

  8. Run the deterministic checker:

    ./.venv/bin/python .agents/skills/manor-workspace-blueprint/scripts/check_blueprint.py path/to/blueprint.json
    

    Compare source and installed exports with the semantic round-trip checker:

    ./.venv/bin/python .agents/skills/manor-workspace-blueprint/scripts/compare_roundtrip.py \
      source.json installed.json
    
  9. Add or update a regression test for every parity rule. At minimum run the relevant tests/test_blueprint_v11_e2e.py, exporter/installer tests, and git diff --check.

  10. Only after the runtime projection round-trip and the separate one-shot install checks pass may the Blueprint be published, seeded as built-in content, or used as an upgrade source.

Failure policy

Stop and report the exact object and boundary when any of these occur:

  • exported content cannot be installed;
  • installed content cannot be exported back;
  • a Workflow step or dependency changes;
  • an entity-level Task category, SLA, or SLA-linked escalation policy is placed in a Workspace Blueprint;
  • an Automation target resolves to a different Workflow/Agent;
  • an explicitly portable configuration is dropped;
  • an artifact/output contract changes;
  • approval or governance scope becomes broader or narrower;
  • human-readable Blueprint copy disagrees with the executable contract.

Do not repair these by adding another conflicting prompt. Fix the shared exporter/installer contract or remove the unsupported declaration.

Signals

GitHub stars
171
Forks
52
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
manor-workspace-blueprint
Source
github.com/manor-os/manor-ai