Evidence-bounded cultivation operations for lawful cannabis teams.
CanopyOps is an installable AI skill that turns room data, crop observations, logs, and facility constraints into reviewable crop plans, incident workups, transparent calculations, harvest-readiness reviews, CAPA, runbooks, and operating records—while keeping evidence, uncertainty, ownership, and human authority visible.
Open the project site → · Start here · Install · Choose the correct release
- Plans rooms and crops from facility limits, crop stage, cultivar information, measurement methods, targets, labor, and unresolved decisions.
- Works incidents without pet theories by preserving competing explanations and separating reversible containment from cause-specific correction.
- Calculates transparently across VPD, DLI, irrigation, runoff, dryback, EC, pH, and normalized units.
- Reviews harvest and quality evidence without confusing schedule pressure, interpretation, holds, recommendation, and release authority.
- Produces operating memory through incident reports, CAPA, risk registers, room runbooks, crop walks, drying logs, cultivation decisions, and shift handoffs.
CanopyOps does not flatten every sentence into “the AI says so.” It preserves nine distinct states:
observed → measured → calculated → assumed → interpreted → recommended → approved → executed → verified
A supplied target remains a comparison value until its source, crop stage, scope, tolerance, measurement method, and accountable approval establish an active target. A recommendation never becomes authorization merely because it sounds polished.
Facility SOPs, emergency procedures, current approved labels, current jurisdiction sources, qualified laboratory evidence, equipment documentation, and accountable humans remain above the model.
Install CanopyOps, start a fresh task, and use fictional or sanitized data:
Use CanopyOps for a fictional licensed cultivation facility. A late-flower room held 27 C and 78% RH for 42 minutes overnight. One wall sensor recorded the excursion; its calibration status and exact canopy position are unknown. The active target is 25 C and no more than 60% RH, but I have not supplied the approved source or tolerance. Build a provisional incident workup and incident record. Preserve competing explanations, distinguish reversible containment from cause-specific correction, identify the evidence that would change the next decision, and do not claim that any setting was approved or changed.
A useful result should expose missing measurement context, avoid treating one sensor as whole-room proof, preserve plausible alternatives, propose only bounded next steps, and end with an explicit status such as provisional or awaiting authority.
The complete path is in JUDGE-QUICKSTART.md.
CanopyOps has two deliberately separate package lines. Do not mix their files.
| Line | Version | Best for | What is established |
|---|---|---|---|
| Repository-native source and plugin | v0.1.5 | Direct GitHub marketplace installation, root source inspection, root tests, and the branded plugin | The v0.1.5 verification and custody records apply at their stated dates and scopes |
| Settled portable bundle | v0.1.6 | One self-verifying archive with Codex and Claude payloads, detached checksums, and package-specific docs | Static package structure, archive membership, byte parity, and custody—not host activation or field behavior |
Read RELEASE-STATUS.md before choosing an artifact. It is the canonical answer when version labels or platform states appear to conflict.
For the repository-native Codex plugin:
codex plugin marketplace add Stunspot/CanopyOps
codex plugin add canopyops@collaborative-dynamics
Start a fresh Codex task after installation. Portable v0.1.6, standalone Codex, Claude.ai, Claude Code, updating, removal, and recovery paths are documented in INSTALL.md.
CanopyOps works primarily in readable Markdown, CSV, and JSON. Included templates cover:
- facility and crop profiles;
- crop plans and room runbooks;
- incident reports and CAPA;
- cultivation decisions and compliance-verification briefs;
- risk registers, crop walks, drying logs, and shift handoffs;
- harvest-readiness reviews with explicit holds and authority.
See EXAMPLE-TOUR.md for four complete fictional demonstrations.
The repository includes a 20-test deterministic suite covering calculations, validation, package parity, version custody, documentation reachability, release-story consistency, and Pages-local assets. The historical v0.1.5 verification record documents the checks executed against that release candidate. The v0.1.6 bundle carries its own portable verifier and package evidence.
These are different claims:
- a file exists;
- a package validates;
- a host discovers the skill;
- an invocation reaches the intended workflow;
- the tools are healthy;
- the behavior is useful;
- a cultivation decision is correct;
- a human is authorized to act.
CanopyOps does not collapse them. No field pilot, broad reliability guarantee, current jurisdiction coverage, legal opinion, pesticide authorization, batch-release authority, direct equipment control, or customer outcome is claimed.
- Current release and platform status
- Current documentation status
- v0.1.5 verification record
- Safety and scope
- Data and privacy
- Security
CanopyOps was conceived and built on July 17, 2026, during the OpenAI Build Week submission period. Stun supplied the product intent, source material, Ella Greenfield persona, domain and authority boundaries, evaluation philosophy, and release judgment. Codex with GPT-5.6 turned that direction into the routed skill, deterministic utilities, schemas, templates, evaluations, host adapters, documentation, licensing, packaging, verification, and public repository.
The initial working Augment emerged in roughly an hour; public packaging, branding, hardening, documentation, and release custody continued afterward. Read BUILD-WEEK.md for the architecture, provenance, and human/AI responsibility split.
canopyops/— repository-native v0.1.5 skill tree.plugins/canopyops/— branded repository-native Codex plugin.claude-ai/— repository-native Claude.ai upload archives.releases/v0.1.6/— settled portable v0.1.6 bundle, checksums, manifest, receipt, and docs.tests/— deterministic repository checks.docs/— GitHub Pages source and generated raster hero.verification/— retained evidence and review custody.release-assets/v0.1.5/— governed v0.1.5 release objects and custody records.
CanopyOps is a Collaborative Dynamics Augment created by Sam Walker (stunspot), with the Ella Greenfield cultivation persona operating inside an evidence-bounded lawful-market system.
The authentic unmodified authored Augment uses CC-BY-ND-4.0; Python scripts, tests, and machine-readable schemas use MIT. Names and marks remain governed by the trademark policy.