Lawful cannabis cultivation · evidence before certainty
CanopyOps is an installable AI skill for cannabis cultivation leaders, agronomists, crop steering teams, quality staff, and compliance operators. It structures room data, observations, facility constraints, and unresolved questions into reviewable plans, incident workups, calculations, CAPA, runbooks, and handoffs.
Advisory decision support · Human authority stays explicit · No direct equipment control
Product fit
Use CanopyOps when the job is to organize evidence, reproduce a calculation, preserve competing explanations, name an approval gate, or hand a decision to the next accountable human.
CanopyOps preserves nine states: observed → measured → calculated → assumed → interpreted → recommended → approved → executed → verified.
Current delivery · v0.3.1
The portable crop workspace keeps observations, comparisons and records together. This update adds product icons and optional desktop entries. A plugin is not required.
Historical native plugin and earlier portable releases retain their original identity. Native Mac launch, fresh-host activation and field outcomes remain unverified.
Installation
Keep operational records outside the product directory. Extract into a new directory and keep your existing record store when updating.
LOCAL CROP DESK
Download and extract the current ZIP. With Python 3.10+ installed, run python tools/verify_release.py . from the extracted root. Require "ok": true and no findings.
Open the launcher under codex/canopyops/skills/canopyops/workspace/. The workspace guide explains choosing your external record store and recognizing a successful opening.
OPTIONAL DESKTOP ENTRY
The included workspace/create_entry.py creates a Windows shortcut or portable Mac app using an existing desktop directory, an explicit external store and a positive port. It refuses to replace an existing entry. Follow the complete workspace procedure before creating it.
CODEX · COMPLETE SKILL
Copy the entire extracted codex/canopyops/skills/canopyops/ directory into your host's configured skill location. Keep all supporting folders beside SKILL.md, then start a fresh task and check discovery.
CLAUDE · COMPLETE SKILL
The bundle includes claude/canopyops-v0.3.1.zip. Follow its Claude installation guide and your host's current controls. Upload, discovery, invocation and successful work are separate observations; live activation remains unverified here.
Verify discovery, then reach first value
Start a fresh task or conversation after installation. Explicitly invoke CanopyOps if the host supports it, then keep discovery evidence separate from output quality.
DISCOVERY CHECK
Use CanopyOps to outline the evidence you would need before reviewing a fictional late-flower cannabis humidity excursion. Do not diagnose or recommend operational changes yet.
A discovered CanopyOps skill should request facility context, crop stage, duration, sensor location and method, approved-target source and tolerance, observations, prior conditions, equipment state, and accountable authority.
FIRST COMPLETE WORKUP
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 record. Preserve competing explanations, separate reversible containment from cause-specific correction, identify evidence that would change the next decision, and do not claim that any setting was approved or changed.
Representative workflows
Inputs may be prose, tables, logs, CSV, JSON, approved documents, measurements, images supplied for interpretation, or existing operating records. Outputs are drafts for inspection, not silent execution.
Inputs: facility limits, crop stage, cultivar evidence, targets, labor, measurement methods.
Outputs: crop plan, room runbook, risk register, open-decision list.
Inputs: observations, logs, spatial pattern, timeline, equipment state, prior actions.
Outputs: incident workup, evidence gaps, containment, decision record, CAPA.
Inputs: temperature, RH, leaf temperature, light, irrigation, substrate, runoff, EC, pH.
Outputs: transparent calculations, comparison tables, method caveats, next measurement.
Inputs: readiness observations, sample methods, drying conditions, laboratory evidence, holds.
Outputs: harvest-readiness review, drying log, hold list, release-preparation brief.
Inputs: facility SOPs, labels, current jurisdiction sources, named owners, approvals, deadlines.
Outputs: compliance-verification brief, cultivation decision, crop walk, shift handoff.
Inputs: completed artifact, evidence ledger, unresolved conditions, approval and verification records.
Outputs: review findings, status, blocked conditions, named next owner; never automatic release authority.
Configuration and methods
Ordinary reasoning requires no CanopyOps account, API key, or product configuration file. The meaningful configuration is the operating context you supply: facility, room, crop stage, units, sensor method and location, time window, approved targets and tolerances, applicable SOPs, jurisdiction, owners, and authority.
Optional for ordinary reasoning; required for bundled deterministic utilities and the portable v0.1.7 verifier. Repository-native utilities use the Python standard library.
Facility SOPs, emergency procedures, current labels, current jurisdiction sources, laboratory evidence, and equipment documentation outrank model memory.
Keep facility and crop records in an approved workspace outside the installation tree. CanopyOps does not create backups or manage retention for you.
Troubleshooting, recovery, and cleanup
Confirm the final path ends in canopyops/SKILL.md, the complete package is present, and stale duplicates are disabled. Start a fresh task or restart the host once.
Reinstall the complete matching package. If a script is unavailable, confirm Python 3 and the specific workspace permission; otherwise use the documented manual fallback and label it unverified.
Stop. Compare the detached checksum, discard the failed extraction, extract again into a clean directory, and continue only after the verifier returns no findings.
Identify the installed line, preserve records outside it, replace the complete matching package, start fresh, repeat discovery, and rerun the appropriate verifier. Never overlay v0.1.7 onto v0.1.5.
Remove the plugin through the host, or remove the complete installed skill directory. Disable or delete Claude.ai uploads. Archive or delete extracted packages under your retention policy.
If records were wrongly stored inside the install tree, stop changes, preserve what remains, restore from your already-established backup or version history, and move recovered records outside the package. Never invent missing operational records.
Removing CanopyOps does not delete artifacts stored elsewhere. CanopyOps itself does not back up, synchronize, or clean those records.
Validation and provenance
The current repository suite contains 21 deterministic tests for calculations, validation, package parity, version custody, documentation reachability, release-story consistency, and Pages-local assets. The portable v0.1.7 bundle has its own verifier for package structure and byte custody.
The source, plugin, portable bundle, templates, scripts, schemas, examples, and documentation are present.
Repository tests and the portable verifier can check recorded structure, links, relationships, checksums, archive membership, and parity.
Fresh-host installation, host discovery, invocation, tool execution, useful behavior, field fitness, legal correctness, regulatory currency, or customer outcomes.
Historical v0.1.5 evidence remains historical. Portable v0.1.7 evidence applies only to the exact package objects it names. Current documentation review evidence is bound to its recorded content fingerprint.
Support, contribution, and rights
For a reproducible product defect, open a GitHub issue with sanitized inputs, expected and actual behavior, distribution line, version, host, and exact error. For an active alarm or life-safety event, use facility emergency procedures—not repository support.
Read the troubleshooting guide first, then report a minimal sanitized reproduction.
Support policyPreserve evidence states, safety boundaries, source custody, package-line separation, and deterministic verification.
Contribution guideAuthored Augment material uses CC-BY-ND-4.0; Python, tests, and machine-readable schemas use MIT. Marks remain governed separately.
License · Trademarks