Canopy capabilities
UniverLab's catalog of AI agent skills — reusable capabilities for Claude Code and compatible harnesses.
npx -y skills add UniverLab/skills --skill canopy-capabilitiesAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use this skill when someone asks what Canopy can do: onboarding a new user, answering "what can I do here?", proposing automations for a workspace, or deciding whether a need maps to an agent, a loop, memory, or sync. It gives the capability map plus concrete starter ideas, and teaches how to discover the live tool surface instead of trusting a static list.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
4.0 KB, 794 tokens by cl100k_base, as published. Nobody here has run it
Canopy Capabilities: What Can I Do Here?
Canopy is a runtime layer that sits alongside AI coding agents (Claude, Codex, OpenCode, Gemini, Copilot, and any CLI in its platform registry) and gives them what harnesses don't: persistent memory, background scheduling, multi-agent coordination, and autonomous work loops.
This skill is a MAP, not an inventory. Tool names and flags evolve — always verify the live surface (below) before promising an exact call.
Discover the live surface first
get_tools(scope="session_start")— the workspace brief + which MCP tools apply right now.canopy --helpandcanopy <command> --help— the CLI surface.agent_models— which CLI platforms and models this machine can actually run.blueprint_list— ready-made node blueprints for building loops.
If this skill and the live surface disagree, the live surface wins.
The capability map
1. Memory that survives sessions (Project Intelligence Layer)
Facts, patterns, and session summaries persist per project and load into any
future session — across different harnesses. Ask: "what does this workspace
already know?" before re-deriving anything. (Details: canopy-intelligence.)
2. Background agents (cron / file-watch)
Any prompt can run on a schedule or fire when files change — on any installed CLI platform. Starter ideas: nightly dependency audit, docs drift checker, "summarize what changed today", data-pipeline sanity checks on file arrival.
3. Loops (autonomous work queues)
A reusable graph of agent / check / gate nodes consumes a queue of specs:
implement → verify with real commands → review on a different model → commit.
Runs unattended, survives restarts, schedules its own resumption after CLI
quota resets. This is the heavy hitter: backlogs of bugfixes, refactors,
docs, or migrations executed overnight. (Design guidance: canopy-loop-design.)
4. Multi-agent sync
Agents declare missions, report status, and broadcast milestones so parallel
sessions stop colliding in one workspace — and solo sessions leave a legible
history. (Details: canopy-sync.)
5. Interactive session management (TUI)
Live terminals for interactive agents, status semaphores, a prompt builder (with scheduled sends), project views with backlog/knowledge/history, and live loop graphs — one dashboard over everything above.
6. RAG over local documents
Index a directory of PDFs/documents and query it from any agent — useful for literature-driven work (specs that cite real sources) and project docs.
7. Spec backlogs and pools
Work items (specs) live in a standalone backlog, tagged per project, ordered
in pools, and fed to any compatible loop. Writing good specs is a skill:
role / what / how (see canopy-loop-design).
Mapping a need to a capability
| The user says | Reach for |
|---|---|
| "I keep re-explaining my codebase" | Intelligence layer (facts/patterns) |
| "Run X every night / when Y changes" | Background agent (cron / watch) |
| "Work through this backlog without me" | Specs + pool + loop |
| "Two agents keep stepping on each other" | Sync protocol |
| "Ask questions over these papers/docs" | RAG |
| "Send my agent a prompt at 6am" | Prompt builder scheduled send |
When proposing, be concrete: name the trigger, the platform, the verification step, and what "done" looks like. An automation without a deterministic check is a wish, not a capability.
What ships with it: 2 files
1.7 KB alongside SKILL.md