██████╗██████╗ ███████╗    ███████╗██╗  ██╗██╗██╗     ██╗     ███████╗
██╔════╝██╔══██╗██╔════╝    ██╔════╝██║ ██╔╝██║██║     ██║     ██╔════╝
██║     ██████╔╝█████╗      ███████╗█████╔╝ ██║██║     ██║     ███████╗
██║     ██╔══██╗██╔══╝      ╚════██║██╔═██╗ ██║██║     ██║     ╚════██║
╚██████╗██║  ██║███████╗    ███████║██║  ██╗██║███████╗███████╗███████║
 ╚═════╝╚═╝  ╚═╝╚══════╝    ╚══════╝╚═╝  ╚═╝╚═╝╚══════╝╚══════╝╚══════╝

        v5.2.1  ·  MCP reconnect patch — plugin-root entrypoint, Claude/Codex cache compatibility

                                │                                                                         |        
            *                   │           │           │                     _                          /|     |#|
           /|                  _│_         _|_          A          _        _[_]_                _|_    / |     |#|
          /_W_                /:::\       [_=_]        /_\        | |     _/_:::_\_      __     |:::|  |  |     |#|
          ( • )              /|:::|\      |:::|       ((^))       | |    |X/:::::\X|    |__|    |:::|  |  |     |#|
        \__/ \__[]           | |:::| |   _|_:_|_     /(:::)\     _|_|_   |/X:::::X\|    |::|    |:::|_ |  |     |#|
           | |              | |:::| |   |:::::::|    |:::::|    |:::::|  |\X:::::X/|    |::|    |:::::||  |    _|#|_
          /   \             | |:::| |  _|:::::::|_   |:::::|   _|:::::|_ |X/:::::\X|   _|::_|   |:::::||::|   |:::::|
         _|___|_            | |:::| | |:::::::::::|  |:::::|  |:::::::::||/X:::::X\|  |::::::| _|:::::||::|  _|:::::|_
    ___|=======|___________|_|:::|_|_|:::::::::::|__|:::::|__|:::::::::||\X:::::X/|__|::::::||:::::::||::|_|:::::::::|
           BROOKLYN BRIDGE                           WILLIAMSBURG BRIDGE                     QUEENSBORO BRIDGE
      _|_                     _|_                  |▔▔▔|               |▔▔▔|               _|_         _|_         _|_
     | ∩ |                   | ∩ |                 |XXX|               |XXX|              //|\\       //|\\       //|\\
    /| ∩ |\                 /| ∩ |\               /|   |\-------___---/|   |\            ///|\\\_____///|\\\_____///|\\
   / |___| \-------_-------/ |___| \-------------/ |___| \     /   \  /|___| \----------////|\\\\   ////|\\\\   ////|\\
  / /     \ \     / \     / /     \ \           / /     \ \   /     \/ /     \ \       / /  |  \ \ / /  |  \ \ / /  |  \
≈/≈/≈≈≈≈≈≈≈\≈\≈≈≈/≈≈≈\≈≈≈/≈/≈≈≈≈≈≈≈\≈\≈≈≈≈≈≈≈≈≈/≈/≈≈≈≈≈≈≈\≈\≈/≈≈≈≈≈≈≈\≈/≈≈≈≈≈≈\≈\≈≈≈≈≈/≈/≈≈≈|≈≈≈\≈/≈/≈≈≈|≈≈≈\≈/≈/≈≈|≈≈≈\
≈/≈/≈≈≈≈≈≈≈≈\≈\≈/≈≈≈≈≈\≈/≈/≈≈≈≈≈≈≈≈≈\≈\≈≈≈≈≈≈≈/≈/≈≈≈≈≈≈≈≈≈\≈\≈≈ EAST RIVER ≈≈≈\≈/≈≈≈≈/≈/≈≈≈≈|≈≈≈≈\≈/≈≈≈≈|≈≈≈≈\≈/≈≈≈|≈≈≈≈\

CRE Skills Plugin

A Claude plugin delivering a large library of commercial real estate skills covering the full investment lifecycle -- deal sourcing, screening, underwriting, structuring, due diligence, capital markets, market research, asset management, leasing, investor relations, development, disposition, tax planning, ESG, portfolio strategy, and daily property operations. Each skill includes structured process logic, reference documents, chain connections to other skills, and Python calculators for precise quantitative output. Deploys as a plugin in Claude Code (CLI or Desktop Code tab) or as a local MCP server for Claude Desktop Chat tab. See Key Stats for current counts and Release Maturity for status by surface.

How the skill taxonomy works for your team

Every skill now declares a classification that tells your team what it is and how much to trust its output. In plain terms: a micro or normal skill does one bounded job — normalize a T-12, size a loan, draft a quarterly LP letter — and is invoked as a single unit. An orchestrator is a conductor that runs several skills in sequence (an underwrite-to-IC pipeline, a document-to-database ingest); it is a workflow, not an atomic tool. A workspace is a business-facing surface that routes a whole domain — your fund team's LP reporting, your asset managers' operations cockpit — to the right specialist skill and holds context across sessions. For a deal team, an asset manager, or a fund team, the classification is not cosmetic: it determines which outputs are decision-grade and therefore need a human sign-off before they reach an investment committee, a lender, or an LP. Skills whose output can land on a board deck or a distribution notice are marked decision-grade and carry an explicit human gate (analyst review, AM/CFO sign-off, IC approval, or lender/investor review) plus a source posture that refuses to emit a final figure it cannot trace to your data — no fabricated model references, no unresolved $X placeholders in a final-marked report. Everything else is advisory screening output your operator validates. See docs/DATA_GRADES.md for the data-grade ladder and docs/integrations/amos-skill-manifest.md for how this metadata is exported to AMOS.

Open core

This repository is the free, open-source core of CRE Skills, licensed under Apache-2.0. It is the knowledge layer that runs locally in your Claude Code session: the skill library, routing, calculators, and the MCP surface. It stays free and open.

A separate, paid edition, cre-skills-pro, is in development as the institution-grade governance layer that composes on top of this core. It adds a deterministic control plane: lifecycle hooks, a four-eyes approval matrix, append-only redacted deal-state memory, an audit log, and a first-run setup TUI, so decision-grade output (IC memos, valuations, lender and LP deliverables) is gated, provenanced, and auditable for institutional use. It will be offered commercially; this free core is not going away.

Anonymous usage signals and any feedback you choose to send remotely also help shape the paid edition. No deal data, financial figures, prompts, or PII are ever collected or used for this. See PRIVACY.md for exactly what is and is not used, and how to keep everything local.

Release Maturity

This release is an internal beta / controlled release candidate. Most top-level skills (1:1 with a single src/skills/<slug>/SKILL.md) are self-contained and usable today. The residential multifamily subsystem and several install surfaces are held to a higher fail-closed bar and are labeled below.

| Surface / component | Status | What this means | |---|---|---| | Top-level skills (all but one) | Deployed | Runnable today. Expected behavior per SKILL.md. | | residential_multifamily subsystem | Stable, awaiting shakedown (v1.0.0-rc1) | Code-complete. Routing core, workflows, roles, tailoring guards (approval-floor / canonical-redefinition / preview-bundle / missing-doc blocker), and refusal-on-missing-input contracts are all active and under test. Subsystem still ships with placeholder org overlays; decision-grade use requires an org onboarding pass (tailoring interview) to supply real data. Final-marked outputs fail closed when required inputs are absent, enforce a period-seal gate (close_status, close_lock_timestamp, budget_version), and must follow the executive output contract. Graduation to status: stable is gated on the first operator shakedown log — see PREVIEW_MODE.md for the contract. | | Orchestrators | Template / semi-manual | Orchestrators are phase + agent + verdict templates. There is no autonomous engine that sequences phases, polls agents, or aggregates verdicts without Claude acting as conductor. Treat as structured prompts, not fire-and-forget pipelines. | | Marketplace install | Supported (CLI) | claude plugin marketplace add from Claude Code CLI. Claude Desktop's "Add marketplace" dialog is not supported — this repo does not expose a manifest at that URL. See docs/WHAT-TO-USE-WHEN.md. | | macOS DMG / Windows EXE installer | Supported | Smoke-tested by scripts/installer_smoke_test.py (fresh install). | | Cowork ZIP import | Partial | Skills + agents + commands only. Hooks, MCP tools, orchestrators, calculators are not part of the Cowork surface. | | Manual MCP config (Claude Desktop Chat tab) | Supported | .mcp.json + mcp-server.mjs; operational MCP tools with organizational aliases. | | Codex / Gemini / Grok / Manus portable ZIP | Experimental (structural CI) | Skills ship as SKILL.md files. A structural smoke test (tests/install_smoke/test_portable_zip.py + the Portable ZIP Smoke workflow) validates ZIP layout, skills-tree mirroring, frontmatter contract, and runtime-file exclusion. Cross-runtime invocation (the skill actually loading and running inside Codex / Gemini / Grok / Manus) is not tested. Treat as experimental. |

Upgrade, uninstall/reinstall, and corrupted-config recovery paths do not currently have automated smoke tests. The existing smoke tests exercise fresh-install only. See docs/install_smoke_test_matrix.md for the full coverage matrix and gaps.

Known Limitations

Full honest statement: docs/known-limitations.md. The headline v5.2.0 limitations:

• Connectors are not live. Every connector contract and vendor adapter is status: stub; there is no live API feed to any system. The only active ingestion shape is a shared-drive / email file drop. See docs/connectors/CAPABILITY-MATRIX.md (CoStar is not-supported-live per its AI-use T&C; Yardi / MRI / RealPage are blocked-by-vendor).
• Generalized cross-skill governance runtime is still deferred. Decision-grade runtime enforcement (source-class tagging, refusal-on-missing-input, period-seal, placeholder scanner) runs as deployed runtime only inside residential_multifamily. Elsewhere the runtime leg is the final_marked selector + the **t