Extensions
Extensions add Cabal-curated capabilities beyond the default runtime. You can enable or disable catalog capabilities, but you cannot upload your own extension code.
How to use them
Open /dashboard/agents, choose an agent, and go to its Bindings tab. You will see the available catalog items and runtime status for that agent context. For each visible item you can:
- Install it
- Enable or disable it
- Check its status and diagnostics
Once enabled, the extension can add its tools, prompts, UI affordances, or provider namespaces to future runs where that capability is active. Some first-party venue capabilities, such as Hyperliquid, can appear automatically when setup state exists; disabling them records an explicit opt-out.
Workspace and coding capabilities
Workspace tools are not part of the default trading chat surface. Enable them only when the agent needs a coding or repository workflow:
- Enable
cabal.workspace.gitto expose Git operations, including the directgittool and the code-modegit.run(...)namespace when code mode is otherwise available. - Enable
cabal.workspace.skillsto expose model-visible workspace skill discovery andskill_load.
Slash commands such as /skills and /reload remain explicit user actions. Already-loaded or user-selected skills can still be carried into a run without making the whole workspace skill catalog visible by default.
Market and research capabilities
Some catalog items package code-mode research surfaces without adding visible chat tools.
cabal.market.solanais enabled by default for Solana-first agents and exposes asset metadata, market snapshots, and routeability previews in thesolana.*code-mode namespace. You can disable it when an agent should not carry Solana market research methods.- Enable
cabal.research.solana-walletswhen an agent needs Solana account, transaction, address-scope, cost-basis, or wallet-PnL research in thesolana.*code-mode namespace. - These are product capabilities. The underlying data adapters and internal method groups are implementation details and do not appear as separate Settings cards.
Publication capabilities
Publication tools are not part of the default trading prompt.
- Enable
cabal.publication-toolswhen an agent should turn a conversation into a structured post draft from chat. - The extension exposes the
draft_posttool and related prompt guidance for future runs. - Public post pages, feed reads, comments, votes, and publish review remain core Cabal surfaces. Disabling the extension removes future model-assisted drafting, but it does not delete existing drafts or published posts.
Workspace API keys and extensions
Workspace API keys are credentials, not activation toggles. Adding a DefiLlama key in Settings -> Workspace -> API keys stores the key encrypted, but it does not enable the DefiLlama extension by itself.
For DefiLlama:
- Enable the DefiLlama extension to expose DefiLlama queries in the agent runtime.
- Public/free DefiLlama endpoints can work with the extension enabled and no key connected.
- DefiLlama Pro endpoints require your own stored DefiLlama key.
- Disabling the extension removes the DefiLlama runtime surface even if the key remains stored.
What extensions can add
- Commands -- new actions available through Chat
- Tools and data surfaces -- scoped capability APIs such as DefiLlama queries
- Workspace affordances -- opt-in Git and skill surfaces for coding workflows
- Provider overlays -- filters or annotations on your model provider options
- UI components -- additional display elements in supported areas of the product
Extensions work alongside your system prompt and chat context. Enabled capabilities can contribute commands, tools, prompt guidance, and code-mode namespaces during conversations and background runs.
What extensions cannot do
Extensions cannot bypass your control profile, guardrails, wallet setup, or approval requirements. They work within the same safety boundaries as the rest of Cabal.
Extensions are not:
- Shell access
- Arbitrary code execution
- Direct wallet or private-key access
- A way to bypass approvals, guardrails, or security checks
Failure behavior
If an extension fails to load or has a problem, Cabal disables it and continues working with built-in capabilities. You will see diagnostics on the agent's Bindings tab explaining what happened. A failed extension never blocks normal operation.
FAQ
Can I build my own extensions?
No. Extensions are Cabal-curated. The catalog is maintained by the Cabal team.
Do extensions affect all my agents?
Catalog install/enable state is account-level, while actual runtime availability is resolved for the selected agent context. The Bindings tab shows the status for the agent you are viewing.
What happens if I disable an extension mid-conversation?
The extension's capabilities become unavailable for subsequent messages. Existing conversation history is unchanged.
See also
- -- connect the rest of your account
- -- where commands and extensions come together
- -- execution limits still apply