What it owns
The skill classifies deploy, observe, recover, configure, and administer intents, then maps them to the Appaloft operation catalog. Deploy requests land on existing CLI, API, or Web operations.
AI skill
Give AI the Appaloft deploy protocol.
The Appaloft skill is the AI-facing entrypoint. It turns requests like deploy this repo into existing CLI, HTTP/API, and Web operations. The MCP/tool gateway handles tool transport, hosted auth, policy, and audit.
Entrypoint
AI agent
State owner
CLI / API / Web
MCP
public tools + Cloud gateway
Secrets
Reference, never print
Outcome
URL + logs + diagnostics
The install command only installs the skill
Deployment uses safe source inspection and the smallest supported entrypoint
MCP tools must map to the operation catalog and dispatch CommandBus/QueryBus
Outcomes must include live URL, logs, diagnostics, and recovery commands
Where MCP fits
MCP is the tool transport for the Appaloft operation catalog: descriptors, handlers, and server packaging belong in public Appaloft; Cloud adds hosted gateway authorization, policy, audit, and team governance.
Safety boundary
AI must avoid .env files, private keys, tokens, cookies, and unmasked secrets. When credentials are needed, it should reference Appaloft secrets, GitHub Secrets, or runtime environment variable names.
Deploy protocol
The standard path is source inspection, entrypoint selection, existing CLI/API/Web operation dispatch, and an outcome packet with URL, ids, status, logs, diagnostics, and recovery readiness.
Deployment flow
1
Install
Install the Appaloft skill into a skill-capable AI host.
2
Ask AI
Ask the agent to deploy the current repository and return URL, logs, diagnostics, and recovery commands.
3
Verify
On failure, read structured status and recovery guidance before choosing the next step.