alfred

Turn an attached org-design document (a company brief, an "AI staff" plan, a founder's "here's who I want on the team" note) into a real Paperclip company with its agents and a principal login. Use when Sir attaches such a document and says "set this up in Paperclip", "build this company", "spin up these agents", or "bootstrap a Paperclip org from this". Confirm-first: derive the structure, show it, and create NOTHING until Sir says yes.

Drive Paperclip (the AI-agent orchestration plane Sir runs at paperclip.<DOMAIN>) end-to-end through the MCP `paperclip` connector. ~48 first-party tools wrapping Paperclip's REST surface — companies, employees (incl. Alfred-as-an-employee), issues, projects, routines, approvals, comments, governance. Use whenever Sir asks about company state, wants to assign work, check what Alfred-the-employee is doing, file/comment on issues, manage routines, or run governance flows. Use whenever Alfred is acting AS an employee under a Paperclip task assignment (the principal/CEO of the company has tasked Alfred — Alfred answers via comments + issue updates).

Schedule reminders, delegate background work to Alfred-the-text-agent, list scheduled jobs, and inspect Hermes runtime — anything that's about Alfred-the-runtime rather than Sir's vault/calendar/finances. Use whenever Sir asks "remind me…", "research X then tell me…", "what reminders do I have set?", or wants to fire-and-forget a long task on text-mode while you stay on the phone with him.

How to behave on a phone call. Voice persona + tool-usage rules for AgentPhone. The Voice Bridge runs this persona as the OpenAI Realtime session prompt; this file is also the canonical text reference for the openclaw main agent so it stays consistent across channels.

Process raw inbound content (emails, voice memos, notes) into structured Obsidian vault records with proper frontmatter, wikilinks, and file placement.

Read operational vault records and extract latent knowledge into structured learning records with proper frontmatter, wikilinks, and file placement.

Fix vault quality issues — broken frontmatter, invalid values, orphaned records, garbage content.

How to read files (audio, documents, images) that Sir sent to Alfred on any channel — Slack file uploads, Telegram voice memos, MMS audio, generic URLs. Converts file references into transcripts + vault records so Alfred doesn't say "I don't see your file".

How to deliver a message to Sir on a specific channel (Slack DM, Telegram, SMS, voice call, email) using cached contact IDs from KNOWN_CONTACTS.md instead of walking workspace directories. Read this whenever Sir says "send/text/call/email me…".

Assemble and deliver Sir's morning briefing as a continuous narrative — ground in last night's digest, ingest overnight inputs, reason about how matters moved, then write a butler's note. Invoked at Sir's local morning by the chore system. Output is BOTH a vault-persisted record (event/daily-brief-<date>.md) and the Slack message Sir sees at breakfast.

Assemble and deliver Sir's evening digest — a backward-looking close of the day that picks up where this morning's brief left off. Reports per-matter outcomes vs expectations, what's still open, and what tomorrow needs to be ready for. Invoked at Sir's local evening (default 19:00 CET / 17:00 UTC) by the chore system. Output is BOTH a vault-persisted record (event/daily-digest-<date>.md) and the Slack message Sir reads as he winds down.

How Alfred performs container restarts, env-var / credential edits, backup triggers, and other destructive ops — with explicit approval doctrine. The operations in this skill CAN'T be undone or have significant blast radius; read the approval flow before invoking any endpoint here.

Cross-tenant operations for Alfred Prime — the only tenant with federated admin access across the fleet. Use `tenant` to directly read/write another tenant's vault, streams, workflows, schedules, chores, or admin surfaces. Use `ask_alfred` to hand a prompt to the peer's Alfred and get his reasoned answer. Use when Sir names another tenant explicitly or asks you to inspect/modify peer data.

Create, edit, list, or delete user-authored SKILL.md files on Sir's request — the conversational shape of teaching yourself a new behaviour. Use whenever Sir says "make that a skill", "set up a skill for X", "edit my <slug> skill", "delete the <slug> skill", or you've spotted a recurring pattern in his behaviour and want to propose codifying it. Companion to `alfred-chore-authoring` (which sets up recurring Temporal workflows). A skill is what you read when you're invoked; a chore is when you fire on a schedule.

One-shot setup of Sir's Sure finance app for a fresh tenant. Connects Sir's bank-feed (manual, by Sir, in the Sure UI), then auto-creates a working baseline of categories, family merchants, tags, rules, and a budget — so that from the next bank-feed sync onward, every transaction is auto-categorised. Use when Sir says "bootstrap my finances", "set up Sure", or first asks Alfred to look at his money. Runs an LLM-driven iterative clustering pipeline that typically reaches 70–80% rule coverage; Sir closes the rest with a few rounds of manual confirmation.

Read net worth, accounts, transactions, categories, and merchants from Sir's self-hosted Sure personal-finance app, log manual transactions and reconcile illiquid asset valuations, and trigger family-wide bank-feed syncs. Use when Sir asks about money (net worth, spending, balances), when generating Money Day briefs, when logging cash movements, or when setting up Lunchflow as a bank-feed provider.

Read, write, search, and reason over the user's personal vault (matters, errands, instincts, chores, events, people, orgs, notes, and 20+ other record types). Use whenever Sir asks about what's in his vault, wants to create/update/delete a record, or wants you to pull context from prior entries.

Create new chores or edit existing ones on Sir's request. A chore is a recurring (or one-shot) Temporal workflow Sir has asked you to set up. This skill teaches the full lifecycle — design the right pattern (agent-driven vs Python-pipeline), author the workflow, register it, test it, and iterate. Use whenever Sir says "set up a chore to…", "every morning remind me to…", "automate X", or asks you to change an existing chore.

Inspect, trigger, pause, and reason about Sir's scheduled chores. Chores are recurring Temporal workflows — some shipped as platform built-ins, some bespoke Python generated for Sir during onboarding. Use whenever Sir asks about what's running on a schedule, wants to manually trigger a chore, pause/unpause one, or check why a chore hasn't produced results.

Conversational management of Sir's Composio-connected third-party apps — list, connect (OAuth + API key), reconnect on expiry, disconnect, and inspect capabilities. Covers the full OAuth lifecycle including the 1h grace window after reconnect.

How to respond (or not respond) to inbound email on Sir's Alfred inbox. Covers reply vs reply-all vs forward, context assembly, authorized senders, and the channel vs stream distinction.

Look inside Alfred's own learning system — what observations have been recorded, what instincts have been learned, what reflections have been synthesized, and what's queued. Use when Sir asks questions like "what have you noticed about me lately?" or "why are you routing X emails to urgent?" or "show me what you've learned this week".

Inspect the operational health of Sir's Alfred instance — running containers, system info, recent activity, worker status, available models, credentials, and container logs. Use when Sir asks "is everything running", "what model are you on", "show me recent activity", or when a chore or workflow is silently failing and you need to diagnose.

Read Plane state (issue details, comment threads) and aggregate cross-channel context (Slack, Plane, Email, Voice, SMS, Omi) for situational awareness. Use when Sir asks what's open in a Plane project, when @alfred is mentioned mid-thread on a Plane issue, or when Sir asks what happened across channels in some recent window.