Spora Core Laravel Package

What: A portable, zero-configuration AI agent orchestration platform that runs on any shared host (cPanel/FTP) without system dependencies beyond PHP 8.4+.

When the user asks to create, set up, scaffold, or configure a new Spora agent, sub-agent, or specialised assistant; OR when a sub-task needs a toolset different from the calling agent's; OR when the user wants to edit an existing agent's configuration (name, description, system_prompt, scheduling, flags). Use for tasks like 'create me a weather agent', 'I need a research sub-agent', 'scaffold a translator', or 'rename agent 42 to Translation bot'. Do NOT use for editing the current agent's notes — those go through write_notes. Recommended tools: agent (operations create_agent, configure_tools, read_agent, update_agent, list_agents, get_available_tools, read_notes, write_notes, write_notes_overwrite).

How to operate the `agent` tool once you have it. Covers: what each of the nine operations does and which are enabled without the operator opting in; adding a skill to your own allowlist via configure_tools settings; the `skills` block on get_available_tools and why `allowed` and `visible[].active` answer different questions; the notes trio (read_notes / write_notes / write_notes_overwrite) and why wholesale replacement is a separate operation. Trigger on: 'what can you configure', 'enable the X tool for yourself', 'add the Y skill to yourself', 'change your own notes', 'which tools do I have', 'what skills can I read'. For BUILDING a new agent (create_agent → configure_tools → read_agent), use the agent-creation skill instead — this one is about operating the tool, not about the new-agent protocol.

Search, retrieve, embed, share, and produce media assets (images, audio, video, documents) from the media library, and organise them in folders. Trigger on any of: 'show this asset', 'render to PNG', 'list derivatives', 'fetch the source code', 'mint a share link', 'convert the derivative', 'what files do we have', 'media archive search', 'file this in a folder', 'create a folder for these documents', or any request that mentions an `asset_id` from a prior media tool call. The skill explains which operation fits the intent, how returned `asset_id`s chain into follow-up calls, the size/mime constraints that decide between inline-bytes and public-URL retrieval, and how the folder tree is shaped.

Create, list, update, delete and trigger scheduled runs (cron or one-shot) and prompt templates attached to this agent. Trigger on: 'schedule this', 'remind me tomorrow at 8', 'set up a daily check', 'every Monday at 9am', 'save this prompt as a template', 'what's on my schedule', 'trigger the X job now'. When a schedule fires later, the prompt is the only task-specific input — the agent's system prompt and tool allowlist are inherited from the agent row, but there is no operator, no chat history, no attachments, and no live state at fire time. The prompt must be self-contained end-to-end: goal, data sources to consult, output format, and what 'done' means.

Delegate work to another agent: `sub_agent` (default — spawns a child task and wait for its answer), `sub_agent_continue` (follow up on a sub-agent task you already started, and wait for the new answer), or `handover` (the target works in the background; you get no answer back). Trigger on: 'delegate this', 'spawn a sub-agent', 'hand off to N', 'let N take it from here', 'send it to N', 'use the N agent for this'. The critical rules: (a) `sub_agent` is the default — only `handover` when the source agent does NOT need to hear back; (b) for `sub_agent` and `handover` the `prompt` is the ONLY input the target receives — source history and inferred context are NOT carried over, so the prompt must be self-contained; a `sub_agent_continue` prompt is the opposite, the target already has that conversation; (c) the parent reads the child's `final_response` as text on the next tick — if the child put an `asset_id` into that text, the parent chains it into a follow-up tool call.

Add or subtract durations from the current time, or convert an epoch to a human-readable datetime in a chosen IANA timezone. Use when the user asks 'what time will it be in 3 hours', '2 hours ago', 'how many minutes until 5pm', or any other arithmetic on the current time. Covers seconds, minutes, hours, days, weeks, and combinations.

Each example shows the exact tool calls and the intermediate results. The agent should follow the same shape: query time → reduce to seconds → apply to epoch → format.