CLI-Anything: Property Meld

A CLI-Anything harness for Property Meld — the first PM work order CLI for AI agents.

Installation

git clone https://github.com/noogalabs/cli-anything-pm.git
cd cli-anything-pm
pip install -e .
playwright install chromium  # for browser backend commands

Post-merge install refresh (operator side)

Editable pipx installs (pipx install --editable) do not auto-update on source pulls. After pulling new commits, run:

pipx reinstall cli-anything-pm        # pm-stable surface
pipx reinstall cli-anything-pm-dev    # pm-dev surface (if installed separately)

The reinstall picks up any CLI-shape changes (new subcommands, new flags, group→command refactors).

This catches the install-lag class of bug, where an operator's pm binary is missing recently merged subcommands or flags. Source is correct; the editable install never refreshed.

Triage tip: If you see command not found or no such option for a CLI command that exists in main HEAD, suspect install lag first. Run pipx reinstall cli-anything-pm before scoping a source bug.

Configuration

Copy the tracked synthetic example to a private local path:

cp config/propertymeld.example.json ~/.claude/credentials/propertymeld-config.json
export PROPERTYMELD_CONFIG=~/.claude/credentials/propertymeld-config.json
export PM_CLIENT_ID=your-client-id
export PM_CLIENT_SECRET=your-client-secret

Set multitenant_id, nexus_account_id, and credentials_path in the private JSON file. Missing or malformed routing fails closed when an action runs; all command help and the runtime command index remain available without config.

Get API credentials from: Property Meld > Settings > API / Nexus API

Quick Start

pm probe                                     # Verify setup
pm work-orders list --status open --json    # List open work orders
pm work-orders get 900001 --json             # Single work order
pm work-orders comments 900001 --json        # Comments (browser)
pm work-orders assign-tech --work-order-id 900001 --tech Tech A --json
pm work-orders assign-vendor --work-order-id 900001 --vendor "Fixture Service" --json
pm vendors invite --email vendor@example.com --first-name Fixture --last-name Vendor --company "Fixture Service" --line1 "123 Main St" --postcode 12345 --phone 2025550110
pm tenants invite --unit-id 9000025 --first-name Fixture --last-name Resident --email resident@example.com --cell 2025550110
pm tenants edit-contact 9000026 --cell 2025550110 --primary-email tenant@example.com
pm index --json                              # Runtime-derived command catalog

Read-only Insights analytics

pm insights melds --limit 100
pm insights turnovers --project --limit 100
pm insights benchmarks --work-category TURNOVER --limit 100

Insights commands fetch only the fixed authenticated Parquet GET endpoints and emit a safe JSON projection. Meld and turnover rows join vendor_assigned_name to the complete Nexus vendor roster. Each row retains the source name and reports resolved, unresolved, ambiguous, or not_applicable; unresolved and ambiguous rows are never discarded. Session expiry fails closed instead of invoking the write-capable recapture path.

Exhaustive work-order reads

--complete is an opt-in read protocol. Existing list/get output shapes stay unchanged without it. Comments, work entries and each uploader-role file source now follow their pagination chains even for legacy list output.

pm work-orders list --complete --limit 1
pm work-orders get 900001 --complete
pm work-orders comments 900001 --complete
pm work-orders files 900001 --complete
pm work-orders work-entries list 900001 --complete
pm work-orders notes 900001 --complete

The first command explicitly ignores --limit and exhausts the list. Its scope is the meld roster, not every child resource on every meld. A per-row work-entry marker still directs callers to the dedicated read. Cookie-path client filters run after source exhaustion; source_count and source_returned describe the pre-filter source, while count describes the filtered result. Existing unsupported filter combinations still refuse. This change adds no updated-since option and makes no claim that such a server filter works.

Collection envelopes use schema_version: 1, results, count, next, complete, pages, returned, basis, and resource. Resource identity binds the backend, endpoint, meld ID where applicable, and query filters. An absent server count remains null; a terminal declared chain can still be certified. A plain array or a payload lacking both a count and an explicit terminal next does not prove completion and is refused in --complete mode. Provenance is reported rather than assuming an undocumented array pagination convention.

Malformed links, changed filter scope, cross-origin/other-endpoint links, cycles, repeated pages/IDs, a 50-page safety cap, inconsistent counts and a terminal row count mismatch fail with a structured nonzero error. Next links are checked before another authenticated GET. Only cursor, offset, page and limit may change between pages; other filters, including repeated status values and the comments meld ID, remain bound. No partial result is certified.

get --complete returns result (the meld, with work_entries), comments, files, notes, and individual resources metadata for meld/notes/comments/ files/work_entries. All requested resources must succeed. Files preserve role identity and per-role counts; identical numeric IDs in different uploader-role tables remain separate files. notes reads existing maintenance_notes and completion_notes fields plus comments and work-entry text, not a new notes endpoint. Reuse these fields from get --complete instead of fetching notes again. Missing note fields or unreadable child pages refuse completeness.

Completion means the reported endpoint chain was exhausted consistently. It does not claim a transactionally frozen snapshot while remote records change. All regression fixtures are synthetic; no live response bodies are test data. Opt-in complete reads fail closed on session expiry without invoking session recapture, a browser or an authentication helper; legacy auth behavior stays.

Architecture

Dual backend:

• API backend (api_backend.py) — Nexus API OAuth2 for all reads
• Browser backend (browser_backend.py) — Playwright for actions API doesn't support

Contributing

This is a CLI-Anything harness. Follow the CLI-Anything contribution guide for CLI-Hub submission.

Building the wheel

Build release wheels from the repository root with:

python setup.py bdist_wheel

The build command recreates both its source-copy staging directory and its final wheel payload directory before copying modules. Generated build/ and dist/ trees are ignored and must not be committed. The test suite verifies that the complete cli_anything/ wheel payload contains exactly the declared Python source members with byte-for-byte parity and no extra file type. It then installs the wheel in a fresh virtual environment and exercises the public pm insights command tree.