Activated Cloud

Builds clean Excel workbooks (.xlsx) with openpyxl: an about sheet, labelled inputs, raw data as a table, and live formulas for every calculation, with proper number formats, frozen headers, validation and a native chart, then recalculates a copy to prove there are no formula errors. Use when the owner wants a spreadsheet, tracker, budget, model or report in Excel or Google Sheets. Not for cleaning or analysing messy data in Python (see clean-and-analyse-data) or for chart images (see make-clear-charts).

Plans and builds a PowerPoint deck (.pptx) that tells one story: storyline first as a list of action titles, then one message per slide with the chart, table or few words that prove it, speaker notes, sources and slide numbers, built with python-pptx (on the owner's template when there is one) and checked by rendering every slide. Use for board updates, pitches, proposals, reviews and training decks. Not for long-form documents (see write-structured-report) or standalone chart images (see make-clear-charts).

Makes finished PDF files on your own computer: reports and letters from HTML and CSS printed with the installed Chromium (or WeasyPrint), data-heavy layouts such as invoices and stock lists with reportlab, and merging, splitting, rotating or filling existing PDFs with pypdf, then renders pages to images to check them. Use when the owner wants a PDF to send, print or archive. Not for an editable Word file (see make-word-document) or for reading a PDF someone sent (see summarise-long-documents).

Builds a clean Word file (.docx) with python-docx on your own computer: real heading and list styles, tables, images, page numbers and comments, or fills and edits an existing template, then checks the file opens and looks right. Use when the owner wants a Word document, a .docx to edit or send, or a template filled in. Not for deciding what the document says (see write-structured-report) or for a fixed-layout PDF (see make-pdf-document).

Reads long documents (PDFs including scans, Word files, slide decks, spreadsheets, long web pages and email threads) all the way through and produces an accurate summary, a question-led answer or an extract of obligations, dates and numbers, with page references for every claim. Use when the owner says read this, summarise it, what does it say about X, or pull out the deadlines. Not for writing a new document from your own material (see write-structured-report) or for researching across the web (see web-research-with-sources).

Writes and debugs Dockerfiles and Compose stacks that build fast, run as non-root, stay small and behave in production: multi-stage builds, pinned base images, cache-friendly layer order, a strict .dockerignore, no secrets in layers, exec-form commands, health checks and readiness-aware Compose dependencies. Includes a method for containers that exit, cannot be reached, lack permissions or fill the disk. Use when asked to containerise an app, fix a broken image or compose file, or shrink an image. Not for rolling releases out to servers (use deploy-and-rollback).

Cleans messy exports (CSV, Excel, copied tables) and analyses them with pandas: profiles every column, fixes types, dates, money and duplicates with a logged, reversible record, joins safely, then answers the owner's question with totals, trends and comparisons that reconcile to the source. Use when the owner says look at this data, what does it tell us, tidy this list, or why did X change. Not for building a formula workbook for others to maintain (see build-excel-spreadsheet) or for styling charts (see make-clear-charts).

Compares options (software, suppliers, plans, venues, hires of a service, ways to solve a problem) with knock-out criteria, weighted scoring agreed up front, true total cost, evidence for every score, a sensitivity check and a clear recommendation with the runner-up and the conditions that would flip it. Use for which should we choose, compare these, or find me the best X for Y. Not for open-ended fact finding (see web-research-with-sources); never signs up, buys or commits without the owner.

Drafts emails in the owner's voice that get the reply they need: a subject that says what and by when, the point or ask in the first line, short body, easy reply, right tone for the relationship, checked names, dates and attachments, saved as a draft in the connected Gmail or Outlook (or the signed-in browser) and sent only on the owner's explicit yes. Use for replies, requests, follow-ups, introductions, declines and tricky messages. Not for internal team updates (see status-updates-and-handoffs).

Gets real tasks done in websites with your own browser: filling forms, downloading reports and invoices, checking dashboards and updating settings on sites the owner has signed you into, step by step with snapshots, stopping for the owner's approval before anything that submits, pays, sends, publishes or deletes, and keeping proof. Use when a job lives in a website with no connected app. Not for reading public pages for facts (see web-research-with-sources, which uses web_extract).

Makes a small, targeted code or config change (a fix, a rename, a default, a copy change, one function) with the smallest correct diff: locate the exact site, read it first, find every caller before touching an interface, match the file's style, edit with patch, validate the syntax, run the relevant checks and show the diff. Use when the change fits in one or a few files and needs no plan. Not for multi-step features (use implement-feature-end-to-end) or behaviour-preserving restructuring (use refactor-safely).

Makes honest, readable charts with matplotlib (PNG and SVG for documents, decks and chat) or as native editable charts in Excel and PowerPoint: picks the right chart for the question, titles it with the finding, highlights what matters, labels directly, and checks the image by looking at it. Use when numbers need to be shown, not listed, or the owner asks for a graph. Not for working out the numbers (see clean-and-analyse-data) or building the workbook around them (see build-excel-spreadsheet).

Builds a working map of a repository you have not worked in: what it does, how it is laid out, the exact commands to install, run, test, lint, type-check and build it, a baseline of what already passes, the entry points, one traced request, and the conventions to copy. Use before your first change in a repo, when asked how something works, or when picking up a project. Not for planning a specific change (use plan-code-change) or chasing a bug (use debug-root-cause).

Turns a meeting into something people can act on: before it, a purpose-led agenda with pre-reads; after it, notes from the transcript or the owner's jottings that separate decisions, actions (one owner, one date each), open questions and key facts, distributed only with the owner's OK, with actions tracked and chased. Use for prepare the agenda, write up this call, what did we agree, or who owes what. Not for general status reports (see status-updates-and-handoffs).

Keeps your computer's files findable: one folder per job in your work space with inputs, working and outputs, clear dated and versioned file names, a README per job, the shared Team folder used properly, and tidy-ups done as inventory, approved plan, logged moves you can undo, duplicates found by content, and deletions only to the Trash with the owner's OK. Use for tidy up my files, where did X go, set up a folder for this, or clean out Downloads. Not for scheduling jobs (see schedule-recurring-work).

Researches a question on the web the way a professional fact-checker does: frames sub-questions, searches with variant queries, prefers primary sources, checks who is behind each source, traces claims to their origin, logs every source, and answers with citations, dates and a stated confidence. Use for find out, look into, what is the latest on, or any brief that needs facts the owner can rely on. Not for choosing between products or suppliers (see compare-options-and-recommend) or for reading files the owner gave you (see summarise-long-documents).

Reviews a diff or pull request the way a senior engineer would: understand the intent, check the design fits, read every changed line with its surroundings, verify suspicions by running code, and write findings ranked by severity with file:line, why it matters and a suggested fix, ending in a clear verdict. Covers correctness, security, data safety, tests, operability and performance. Use for a teammate's pull request, your own branch before merge, or code a subagent produced. Not for answering feedback on your own change (use respond-to-code-review).

Sets up recurring and one-off scheduled work with the cronjob tool: agrees what, when (in the owner's time zone) and where results go, writes a self-contained job prompt that stays quiet when there is nothing to say, picks the right options (monitor, continuity, skills, delivery), test-runs it, and keeps the list of jobs tidy. Use for every Monday send me, remind me on, check X daily and tell me if, or on the 1st of each month. Not for work you can do right now (do it with the matching skill, such as get-tasks-done-in-browser), and never for unattended sending, paying or deleting.

Plans and writes a clear document or report: a report, proposal, brief, one-pager, decision memo, policy or how-to, with the answer first, statement headings, sourced numbers and a specific ask. Use when the owner or a teammate wants something written up for someone to read and act on. Not for producing the file itself (see make-word-document or make-pdf-document) or for condensing someone else's long document (see summarise-long-documents).

Writes status updates the owner can read in ten seconds (status first, what changed, what is next, what you need, by when) and handoffs a teammate can pick up without asking a single question (goal, state, files, next steps, decisions, traps). Also covers blockers, escalations and end-of-task reports, sent in chat, to teammates with brief_team or ask_teammate, or to the Team folder. Not for emails to outside people (see draft-emails-for-owner) or meeting write-ups (see meeting-notes-and-actions).

Prepares and publishes a software release: check readiness, collect what changed since the last tag, choose the version number with semantic versioning, write a user-facing changelog and release notes with upgrade steps for breaking changes, bump every version string, tag, publish through the project's pipeline, and verify the published artifact in a clean environment. Publishing and pushing release tags need the owner's go-ahead. Use when asked to ship a version, tag a release or write a changelog. Not for deploying a running service (use deploy-and-rollback).

Designs and analyses A/B tests on pages, emails, ads and flows: an evidence-based hypothesis, one primary metric with guardrails, sample size and run time from the real baseline, randomisation and QA, stopping rules, a sample ratio check, and a plain-English readout with confidence intervals. Also says when traffic is too low to test. Use when asked to A/B test something, which version won, or how long to run a test. Not for finding what to test (use funnel-analysis or landing-page-review).

Tracks paid media spend and returns week by week: pacing against budget, CPA, ROAS and blended MER from the business's own sales data, break-even and payback checks, reconciliation of platform-claimed conversions with real orders or leads, anomaly flags, and clear scale, hold or cut recommendations. Use for a weekly ads report, when asked whether ads are profitable, where to move budget, or why ROAS dropped. Not for building new campaigns (use paid-ads-campaign-plan).

Produces customer case studies end to end: picks the right customer, prepares and runs the interview, verifies every number, writes a results-first story in challenge, approach, results form, and manages the customer's written approval, then cuts it into a one-pager, sales slide and short snippets. Use when asked for a case study, customer story or success story. Not for general articles (use long-form-article-writing) or testimonial requests on social (see the social-media area).

Finds and removes code nothing uses (unused functions, files, exports, endpoints, dependencies, config and finished feature flags) with proof for each removal: static analysis to find candidates, full-text search for dynamic and cross-repo references, git history for intent, runtime evidence where available, and a green build after every batch. Risky removals such as public APIs, endpoints and database columns wait for the owner. Use for codebase cleanups, shrinking bundles, dropping old features or removing a dependency. Not for restructuring live code (use refactor-safely).

Builds and runs an editorial calendar for owned content (blog, newsletter, guides, video, podcast): content pillars and topic clusters, a prioritised backlog, a cadence the team can sustain, a workflow with owners and due dates, and a refresh queue for decaying pages. Use when asked what we should publish, to plan next quarter's content, or to get content organised. Not for per-platform social calendars (see the social-media area) or for writing a piece (use long-form-article-writing).

Diagnoses why marketing or transactional email lands in spam or bounces: checks SPF, DKIM and DMARC with dig and real message headers, the mailbox providers' bulk-sender rules, complaint and bounce rates, blocklists, list hygiene and sending patterns, then writes a fix plan with exact DNS records for the owner to approve. Use when emails go to spam, open rates collapse, or before moving to a new email tool or domain. Not for writing emails (use email-newsletter-production or lifecycle-email-sequences).

Plans, writes, builds and tests a recurring email newsletter: a clear reader promise and format, subject line and preview text, one main story with a single primary call to action, accessible layout, UTM-tagged links, a pre-send QA checklist, and honest reporting that does not rely on opens. Use for a weekly or monthly newsletter, a one-off announcement email or a newsletter relaunch. Not for automated triggered series (use lifecycle-email-sequences) or spam problems (use email-deliverability-check).

Maps a conversion funnel from first visit to purchase or activation, pulls step-by-step numbers from analytics, the CRM or the store, segments them by source, device and cohort, values each leak in money, diagnoses causes with qualitative evidence, and hands back a ranked list of fixes and test ideas. Use when asked where we lose people, why conversions dropped, or what to fix first. Not for designing the test itself (use ab-test-design) or reviewing one page in depth (use landing-page-review).

Takes a feature or multi-file change from request to a verified, committed result: confirm the ask, get a baseline, plan with a todo list, read before every edit, find every caller before changing an interface, build in small tested slices, run the project's own tests, linters and type checks, exercise the feature for real, review your own diff and report with evidence. Use for any change spanning several files or steps. Not for a one-file tweak (use make-surgical-change) or finding a bug's cause (use debug-root-cause).

Reviews a landing page, homepage, pricing or signup page for conversion: checks message match with the traffic source, the five-second clarity of the offer, CTA, proof, objection handling, form friction, speed, mobile and accessibility, scores it with a rubric, and returns quick wins, bigger changes, rewritten headlines and test ideas. Use when a page does not convert, before sending paid traffic to a page, or when asked for feedback on a page. Not for search ranking problems (use seo-site-audit).

Designs, writes and sets up automated lifecycle email sequences: welcome, onboarding and activation, lead nurture, trial-to-paid, abandoned cart, post-purchase, and win-back. Each flow gets a trigger, exit rules, timing, branches and one job per email, and is measured against a holdout group. Use when asked for a welcome series, drip campaign, nurture, onboarding emails or cart recovery. Not for one-off broadcasts (use email-newsletter-production) or inbox placement problems (use email-deliverability-check).

Plans a paid ads campaign on search or social platforms (Google, Microsoft, Meta, LinkedIn and others): objective and target cost per result from unit economics, platform choice, account structure, budget that can exit learning, tracking set-up, creative briefs with a variation matrix, a launch checklist and kill-or-scale rules. Use when asked to plan ads, write ad copy, brief creative, or launch a campaign. Not for weekly spend and ROAS reporting (use ad-budget-and-roas-tracking) or organic social (see the social-media area).

Restructures or simplifies code without changing what it does: pin current behaviour with tests (characterisation tests where coverage is thin), move in small named steps that each keep the build green (extract, inline, rename, move, flatten conditionals, consolidate duplicates), use the compiler, structural search and full-text search to catch every reference, and keep refactors apart from behaviour changes. Use when asked to clean up, simplify, split a large file or prepare code for a feature. Not for removing unused code (use delete-dead-code).

Handles review feedback on your own change with technical rigour: read every comment before acting, restate what each asks, check it against the code, fix valid points one at a time with tests, push back with evidence when a suggestion is wrong or out of scope, reply in each thread, and get CI green before asking for another look. Use when a pull request gets review comments or the owner or a teammate critiques your code. Not for reviewing someone else's change (use review-code-change).

Finds and prioritises the search queries worth targeting using only free sources (Search Console, Keyword Planner, Trends, autocomplete, People Also Ask, forums and customer language), groups them into clusters by shared results, maps each cluster to one page, and labels every volume with its source. Use when asked what keywords to target, what to write about for SEO, or why we get no organic traffic. Not for fixing site problems (use seo-site-audit) or writing the article (use long-form-article-writing).

Audits a website's technical and on-page SEO with free tools: crawls it politely with the bundled script, checks indexing, robots, sitemaps, canonicals, redirects, speed and Core Web Vitals, mobile, structured data, titles, headings, content and internal links, then ranks fixes by impact and effort. Use when asked for an SEO audit, why we are not ranking, why traffic dropped, or before a site migration. Not for choosing topics or keywords (use seo-keyword-research).

Puts every 'done', 'fixed', 'tests pass' or 'deployed' claim behind fresh evidence: name the command that proves the claim, run it in full after your last edit, read the whole result and the exit code, confirm it tested the right thing, then report exactly what it showed and what you could not check. Use before reporting completion, committing, opening a pull request, ticking a todo item or handing work to a teammate. Not a replacement for writing tests (use write-effective-tests).

Finds and fixes the root cause of a bug, crash, wrong output, error log or failing check: build a command that reproduces the exact symptom, read the full stack trace and logs, bisect to the commit or change that broke it, trace the bad value back to where it starts, test one ranked hypothesis at a time, then fix at the source with a regression test and remove the instrumentation. Use for any bug whose cause is not already proven, including 'it worked last week'. Not for sorting incoming reports (use triage-bug-report).

Plans a product or feature launch across every channel: sizes it into a launch tier, writes the launch brief, builds a dated countdown with owners, runs a go/no-go check and a launch-day runbook, then measures against targets. Use when the owner says we are launching X, plan the release, or what do we need for launch day. Not for the social post schedule itself (use campaign-and-launch-planning in social-media) or a year-long plan (use marketing-plan).

Runs a scoped, non-destructive vulnerability assessment of servers, domains and cloud accounts that the owner owns and has authorised in writing in this conversation: inventories exposed services, checks versions, TLS and configuration, and reports ranked findings with fixes. Refuses any target without that written authorisation. Use when the owner asks to check their own infrastructure for weaknesses. Not for source code (use secure-code-review) or web app logic (use web-app-security-review).

Drafts monthly or quarterly investor updates and board packs: headline, key metrics against plan with consistent definitions, cash and runway, highlights, lowlights told plainly, specific asks, and for boards the agenda, papers and resolutions needed. Numbers are reconciled with finance and nothing is sent without the owner's approval. Use before an investor update or a board meeting. Not for internal daily or weekly briefs (use daily-and-weekly-briefing) or one-off decisions (use decision-memo).

Creates a usable brand voice and style guide from the brand's best existing writing and its customers' language: 3 or 4 voice traits with this-not-that rules, a tone map by situation, a word list, mechanics, and before-and-after rewrites. Also audits copy against an existing guide. Use when asked to define our voice, make writers sound consistent, or check copy is on brand. Not for deciding what to say or to whom (use positioning-and-messaging).

Turns finished work into clean history and a reviewable pull request: learn the repo's conventions, branch from an up-to-date base, stage only intended files, write commit messages that explain why, keep commits small and buildable, integrate upstream safely, open a pull request with test evidence, watch CI and undo mistakes without losing work, never force-pushing or rewriting shared history without the owner's go-ahead. Use when committing, pushing, opening or updating a pull request, or tidying a branch. Not for tagging and publishing a version (use cut-a-release).

Maps the competitive landscape for a product or market: direct, indirect and substitute competitors plus the do-nothing option, with positioning, target customers, pricing, strengths, weaknesses and recent moves, all from public sources that are cited and dated, and a positioning map on axes buyers care about. Use for strategy, pitch decks, pricing or product decisions. Not for benchmarking social media accounts (use competitor-analysis in Social Media) or sizing the market (use market-size-estimate).

Writes the owner's scheduled briefings: a morning brief they can read in three minutes (today's meetings with prep notes, top priorities, decisions waiting, deadlines, overnight messages that matter, key numbers) and a weekly brief (results against plan, metrics with change, commitments, risks, next week's shape). Every number is sourced and timed. Use when the owner wants a daily or weekly rundown. Not for investor or board updates (use board-and-investor-update).

Writes a one-to-two page decision memo: the decision and its deadline, who decides and who is consulted, context, criteria, two to four real options including doing nothing, each scored with costs, risks and reversibility, a clear recommendation, the dissent, and how the decision will be checked later. Keeps a decision log. Use when the owner faces a significant choice or asks for a paper for the board. Not for routine updates (use stakeholder-update) or full board packs (use board-and-investor-update).

Audits a project's third-party packages and container images for known vulnerabilities, and scans the code and full git history for leaked keys and passwords, then returns a triaged fix list (reachable, actively exploited, fixed version) and a rotation plan. Use when the owner asks to check dependencies, update vulnerable libraries or find exposed secrets. Not for flaws in the owner's own code (use secure-code-review) or for live servers (use authorised-vulnerability-assessment).

Ships a change to a running environment with a tested way back: pin exactly what is going out and where, record the current version, check migrations and config for compatibility, deploy through the project's established method, verify with real requests, logs and metrics, watch a bake period against rollback triggers set in advance, and roll back fast when they fire. Production deploys and rollbacks need the owner's explicit go-ahead. Use when asked to deploy, release to staging or production, or undo a bad deploy. Not for publishing a library version (use cut-a-release).

Designs or changes an HTTP API so clients can rely on it: resource-oriented URLs, correct methods and status codes, consistent naming, structured problem-details errors, token-based pagination, idempotency keys for safe retries, optimistic concurrency, authorisation on every object, versioning and a strict rule for what counts as breaking, all written into an OpenAPI description with contract tests. Use when adding or changing endpoints, reviewing an API design or planning a new version. Not for debugging calls to an existing API (use debug-root-cause).

Builds an evidence table that lines up every source on a question (what it claims, method, sample, date, reliability and direction) and rolls it up per claim, so agreement, conflict and gaps are visible at a glance and each conclusion shows how much evidence stands behind it. Use when weighing studies or sources for a decision, preparing a review, or when the owner asks 'how solid is this?'. Not for writing the narrative summary (use research-brief) or finding the sources (use sourced-web-research or literature-review).

Checks the specific factual claims in a draft or message (statistics, quotes, dates, names, 'studies show', images) against original sources and returns a verdict per claim with the evidence trail and a suggested correction. Use before publishing a post, deck, report, press release or investor update, or when the owner asks 'is this true?'. Not for open-ended research questions (use sourced-web-research) or reviewing a body of studies (use literature-review).

Makes slow code, pages, queries or services measurably faster: define the metric and target, measure a repeatable baseline, profile to find where the time actually goes (CPU, I/O, database, network, rendering), fix the biggest cost first, and prove the gain with before-and-after numbers. Covers profilers for Python, Node, Go and Rust, SQL query plans, N+1 queries, benchmarking hygiene and Core Web Vitals. Use when something is slow, times out, or costs too much to run. Not for active outages (use incident-first-response).

Diagnoses and fixes a red CI pipeline: find the failing job and the first real error in its log, classify it (test, lint, type, build, dependency, environment, permission, timeout, flaky infrastructure), check whether main fails too, reproduce locally with the same versions and commands, fix the cause and confirm the rerun is green. Also covers writing and editing GitHub Actions and GitLab CI pipelines well. Use when a check is red on a pull request or the main branch. Not for test failures you can already reproduce locally (use fix-failing-tests).

Gets a red test suite green for the right reason: group failures by cause, reproduce one at a time, decide with evidence whether the code, the test or the environment is wrong, fix that, and stamp out flaky tests by finding the race, shared state, clock or ordering problem behind them. Never deletes, skips or loosens a test to get green without the owner's agreement. Use when tests fail locally or in CI after a change, or pass and fail at random. Not for CI failures that are not test failures (use fix-ci-failures).

Triages the owner's inbox to zero two or three times a day: sorts every message into needs-you, draft-ready, delegated, waiting-for, read-later or archive using a rulebook agreed with the owner, drafts replies for approval, spots phishing and payment-change fraud, and sends a short digest. Use when the owner says their inbox is out of control or asks for a daily sweep. Not for scheduling requests (use schedule-and-protect-time) or meeting prep (use meeting-prep-pack).

Runs the first hour of a production incident: confirm and size the user impact, declare a severity (when unsure, pick the higher), get the owner and the right people involved, keep one clear coordinator, mitigate before diagnosing (roll back, restart, scale, disable a flag), keep a timestamped log, post regular status updates, confirm recovery, and set up a blameless postmortem. Risky production actions need the owner's go-ahead unless a runbook authorises them. Use when something is down, broken for users, losing data or alerting. Not for routine bugs (use debug-root-cause).

Runs a structured review of published research (journal articles, preprints, working papers, official reports): a written search protocol, logged screening with inclusion criteria, quality appraisal of each study, an evidence table, and a synthesis that states how strong and consistent the evidence is and where it runs out. Use when the owner asks what the research says, or needs an evidence base for a decision, paper or grant. Not for general web questions (use sourced-web-research) or tabulating sources you already have (use evidence-table).

Researches, outlines, drafts and edits long-form blog posts and articles (roughly 1,200 to 4,000 words) that match search intent, add something the top results lack, and are fact-checked with sources. Use when asked to write a blog post, guide, how-to, explainer or thought-leadership article. Not for customer success stories (use case-study-writing), for choosing topics (use editorial-calendar-planning), or for short social posts (see the social-media area).

Estimates the size of a market as TAM, SAM and SOM using bottom-up and top-down methods side by side, with every input sourced, every assumption labelled, and a low, base and high range instead of one falsely precise number. Use for business plans, pitch decks, pricing, new product or market-entry decisions. Not for mapping who the competitors are (use competitor-landscape-map) or spotting what is changing in the market (use trend-scan).

Builds a quarterly or annual marketing plan that starts from the revenue target, works the funnel backwards into lead and traffic numbers, picks channels on evidence and splits the budget with a test reserve. Use when the owner asks for a marketing plan, next quarter's priorities, or how to spend a budget. Not for a single launch (use product-launch-plan) or for running ads day to day (use paid-ads-campaign-plan).

Turns meeting notes or transcripts into a recap of decisions, actions (one owner, a verb, a due date, a definition of done) and open questions within 24 hours, then keeps a commitments register of what the owner promised and what others promised the owner, with polite chasers and escalation. Use after meetings or when the owner asks what is outstanding. Not for preparing meetings (use meeting-prep-pack) or regular stakeholder reporting (use stakeholder-update).

Prepares a one-page brief before an important meeting: purpose and the outcome the owner wants, who is attending and what each wants, the history with them, key facts and numbers with sources, the agenda, questions and asks, likely objections, and the decisions needed, with pre-reads linked. Use the day before investor, customer, partner, board or hard internal meetings. Not for writing up the meeting afterwards (use meeting-actions-and-commitments) or the board pack itself (use board-and-investor-update).

Scopes a code change and writes a plan another engineer could follow without guessing: goal, testable acceptance criteria, the blast radius (files, callers, data, consumers), a size and risk call, small ordered tasks each with exact paths and a verification command, and a rollback path. Also writes decision records when a change needs a technical choice. Use before multi-file or risky work, anything you will delegate, or when asked for a plan or estimate. Not for doing the work (use implement-feature-end-to-end).

Works out who the product is for, what it beats and why it wins, then writes a positioning canvas and a messaging hierarchy every channel can reuse. Use when the website feels vague, before a launch, rebrand or new market, or when sales and marketing describe the product differently. Not for tone and style rules (use brand-voice-guide) or a channel and budget plan (use marketing-plan).

Analyses qualitative data (interview transcripts, open survey answers, reviews, support tickets, call notes) with thematic analysis: a defined codebook, coded excerpts, themes with how many participants raised each, representative quotes, contradictions and what the themes mean for the decision. Use after interviews or surveys, or when the owner hands over a pile of feedback and asks what it says. Not for designing the questions (use survey-and-interview-design) or tabulating published studies (use evidence-table).

Turns research findings into a decision-ready brief or executive summary: the answer first, three to five key findings with citations, confidence stated in calibrated words, what remains uncertain, and recommended next steps tied to the evidence. Use when the owner says 'summarise this for me', 'write it up for the board', or needs a one-page brief. Not for gathering the evidence (use sourced-web-research or literature-review) or tabulating sources (use evidence-table).

Changes a live database schema or backfills data without downtime or data loss: read the real SQL, classify each statement's lock and rewrite risk, use expand-and-contract for renames and type changes, build indexes concurrently, add constraints unvalidated then validate, batch backfills, set lock timeouts, rehearse on production-sized data and keep a way back, with the owner's go-ahead before dropping data or touching production. Use when writing, reviewing or running a migration (Postgres first, MySQL notes). Not for shipping the application code itself (use deploy-and-rollback).

Runs the owner's calendar by agreed rules: handles meeting requests with two or three time-zone-correct slot options, protects focus blocks and buffers, holds tentatively until confirmed, plans the week ahead every Friday, and audits where time actually goes against the owner's priorities. Use for any scheduling request, a weekly calendar plan, or 'I have no time to think'. Not for inbox sorting (use inbox-triage-to-zero) or meeting briefs (use meeting-prep-pack).

Reviews the owner's codebase or deployment config for security flaws (access control, authentication, injection, unsafe file handling, secrets, crypto misuse, risky config) and returns ranked findings with file and line, impact and a concrete fix. Use when the owner asks for a security review of a repo, a pull request or a Dockerfile, Terraform or web server config. Not for vulnerable third-party packages or leaked keys (use dependency-and-secret-audit) or for checking a running app (use web-app-security-review).

Runs the first hours of a suspected security incident: establishes the facts, sets severity, preserves evidence with a chain of custody, proposes containment for the owner to approve, hunts for the same signs elsewhere, and flags legal notification clocks. Use for 'we've been hacked', a leaked password or key, an odd login, ransomware, a phishing click or a suspicious alert. Not for the write-up once things are stable (use security-post-incident-review).

Hardens the owner's Linux servers, cloud accounts and admin accounts (email, code host, domain registrar) to a recognised baseline: audits the current state read-only, proposes a ranked change list with rollback steps, applies approved changes one at a time, and verifies. Use for a new server or cloud account, or when the owner asks to lock things down. Not for reviewing who has access (use user-access-review) or for backups (use backup-and-restore-check).

Builds or changes user-interface code that fits the existing app: reuse its components, design tokens and data-fetching patterns, handle loading, empty, error and long-content states, keep it accessible by keyboard and screen reader and responsive down to phone width, then verify in a real browser at several widths with a clean console, plus type checks, tests and a production build. Use for UI features, layout fixes and component work in React, Vue, Svelte or plain HTML and CSS. Not for backend endpoints (use design-http-api).

Answers a research question from the open web with every factual claim tied to a cited, dated source: frames the question, plans searches, prefers primary sources, reads laterally to check who is behind each source, triangulates key facts, and labels confidence, thin evidence and conflicts. Use for 'find out', 'look into' or 'what do we know about' questions. Not for academic literature reviews (use literature-review) or checking specific claims in a draft (use fact-checking).

Maps who needs to hear about a project, programme or change (power and interest grid), sets the cadence and channel for each group, and drafts regular status updates with a defined red, amber or green status, progress, next steps, risks and specific asks, delivering bad news early. Use for project updates to partners, customers, leadership or the team. Drafts go to the owner for approval. Not for investor or board reporting (use board-and-investor-update) or meeting recaps (use meeting-actions-and-commitments).

Designs surveys and interview guides that produce usable evidence: research objectives mapped to every question, neutral wording, sound answer scales, screeners, sample size and margin of error, consent and privacy wording, and a pilot before launch. Use when the owner wants to survey customers or staff, run user, customer or expert interviews, or test an idea with real people. Not for analysing the answers afterwards (use qualitative-analysis) or employee engagement programmes (use engagement-survey in People).

Audits and designs a team's rituals (syncs, retrospectives, demos, all-hands, welcomes, farewells, milestones) with a purpose, cadence, format and kill rule for each, and sets up a fair recognition practice: specific peer and manager recognition, an all-hands slot, and a check of who is and is not being recognised. Use when a team feels disconnected, meetings feel pointless, or good work goes unnoticed. Not for surveys (use engagement-survey).

Builds a threat model for a system, feature or vendor integration: a data-flow diagram with trust boundaries, threats found element by element with STRIDE, a ranked risk list, and mitigations with owners and dates. Use when designing or changing a system, before a launch, or when the owner asks what could go wrong. Not for line-by-line code review (use secure-code-review) or testing a live system (use authorised-vulnerability-assessment).

Plans the owner's business trips and events: compares flight, rail and hotel options against preferences and policy, checks entry, visa and passport rules on official sources, builds a day-by-day itinerary with time zones and confirmations, and runs event planning (offsites, dinners, conferences) on a countdown checklist with a budget tracker. Nothing is booked or paid without explicit approval. Use for any trip or event. Not for routine meeting scheduling (use schedule-and-protect-time).

Scans for changes that could affect the owner's business across social, technological, economic, environmental, political and regulatory forces, logs dated signals with sources, clusters them into trends, and rates each for strength of evidence, impact and timing, with indicators to watch. Use for strategy days, annual planning, 'what's changing in our industry?' or horizon scanning. Not for checking a single claim (use fact-checking) or listing competitors (use competitor-landscape-map).

Turns incoming bug reports, error alerts and support escalations, one at a time or as a backlog sweep, into triaged issues: check for duplicates, reproduce on the current version, capture environment and minimal steps, rate severity and priority with a fixed rubric, label and route to an owner, reply to the reporter, and send security reports privately to the owner. Use when new issues arrive or the bug backlog needs a pass. Not for finding the cause or fixing it (use debug-root-cause).

Writes tests that catch real breaks: work test-first with a red-green-refactor loop, name the bug each test guards against, test behaviour through public interfaces, prefer real code and fakes over mocks, derive expected values by hand, cover the edge cases that actually fail, and prove each test can fail. Includes patterns and commands for pytest, Jest, Vitest, Go and Rust. Use when adding a feature, fixing a bug or covering untested code. Not for repairing a red or flaky suite (use fix-failing-tests).

Writes and maintains an account plan for an existing customer: their goals, value delivered so far, stakeholder and relationship map, adoption, whitespace for growth, risks with mitigations, and a quarter's measurable objectives with owners and dates. Use for key accounts each quarter, on handover to a new account manager, or after a big change at the customer. Not for running the renewal or expansion deal itself (use renewal-and-expansion), the customer-facing review deck (use qbr-deck) or researching a prospect (use account-research).

Researches one target or customer account from public sources and the owner's own records, and turns it into a sourced account brief: what they do, their priorities and recent triggers, the buying committee, likely pains as labelled hypotheses, and the best way in. Use before outreach to a named account, before a first meeting, or when planning account-based selling. Not for building lists of many accounts (use prospect-list-building) or a meeting agenda and script (use sales-call-prep).

Checks that the owner's critical data can actually be recovered: inventories what is and is not backed up, agrees recovery point and recovery time targets, compares the setup with 3-2-1 and ransomware-resistant practice, and runs a timed test restore into an isolated place. Use when the owner asks whether backups are OK, after a scare, or for audit evidence. Not for responding to an active incident (use security-incident-triage) or general server hardening (use server-and-cloud-hardening).

Proves the ledger's cash balance agrees to the bank statement at a date: matches every line, lists deposits in transit and unpresented payments, finds errors, and produces a reconciliation that differs by exactly 0.00 with proposed adjusting entries. Use for bank, card, savings, loan and payment-processor accounts at month end or whenever the balances disagree. Not for coding uncategorised lines: use categorise-bank-transactions first.

Builds a driver-based annual budget in a real spreadsheet: revenue from volumes and prices, direct costs from margins or unit costs, a person-by-person headcount plan with employer on-costs, overheads by contract, monthly phasing from seasonality, a cash view, and a checks sheet that proves every total ties. Use when the owner wants next year's budget, a budget to show a lender or investor, or a target to measure monthly results against. Not for the weekly cash view: use thirteen-week-cash-forecast; for what-ifs use financial-scenario-model.

Audits and cleans the owner's CRM so every record can be trusted: data standards, stage definitions with exit criteria, next steps, duplicate detection and safe merges, stale and past-due deals, missing fields, consistent loss and disqualification reasons, opt-out flags and activity logging. Fixes the safe things, asks before merging or changing anything that matters, and never deletes without approval. Use for a weekly or monthly CRM check, before a forecast, or after an import. Not for judging deal health or the forecast (use pipeline-review-and-forecast).

Codes bank, card and payment-processor lines to the owner's chart of accounts with a tax code, a confidence level and a reason for each, matches receipts and payments to open invoices and bills, and proposes bank rules for repeat payees. Use when the feed has uncoded lines, a statement export needs coding, or the owner asks what something should be booked to. Not for proving the bank balance agrees to the books: use bank-reconciliation.

Designs and drafts a multichannel cold outreach sequence (email, LinkedIn and phone) for a researched prospect list: touch plan, personalised copy, reply handling, deliverability and anti-spam compliance (CAN-SPAM, GDPR, PECR and local law), with every message drafted for the owner's approval unless the owner has said to send. Use for first contact with people who have not talked to the owner yet. Not for following up an existing conversation (use sales-follow-up) or building the list (use prospect-list-building).

Assesses how ready the owner is for SOC 2, ISO/IEC 27001 or GDPR: maps current practice to each requirement, marks it met, partial or missing with evidence, and builds a prioritised remediation plan with owners, dates and a realistic path to audit. Use before an audit, a large customer's security review or due diligence, or when the owner asks 'what would it take to get SOC 2?'. Not for writing the policies (use security-policy-writing) or filling in a customer questionnaire (use vendor-security-questionnaire).

Runs discovery and qualifies a deal on evidence: plans and reviews discovery calls, scores the deal with MEDDPICC (or BANT for small, single-buyer deals) using what the buyer actually said or did, finds the gaps, writes the next questions, and recommends advance, hold or disqualify against buyer-verified stage criteria. Use after a first meeting, before moving a deal forward, or when a deal looks stuck. Not for the call prep sheet itself (use sales-call-prep) or price talks (use negotiation-prep).

Drafts or updates a formal HR policy (leave, remote work, expenses, code of conduct, anti-harassment, grievance and disciplinary frameworks, data protection for staff) with scope, rules, procedure, roles, a legal check per country with sources, consultation needs, and a sign-off path through a qualified human. Use when the company needs a binding rule written down. Not for friendly how-to pages (use write-handbook-section) or handling an individual case.

Drafts a job offer from approved terms: a verbal-offer call script, the written offer letter, a summary card for approval, and a checklist of the country's legal points to confirm. Checks every figure against the approved band and flags pay-equity gaps. Use once a hiring decision is made; the owner approves and sends, and a qualified person reviews the template. Not for the full employment contract or for rejections (see screen-cvs-fairly).

Designs, runs and analyses an employee engagement or pulse survey: a short validated-style item set, eNPS, anonymity rules with a minimum reporting group, favourability and driver analysis, theme coding of comments, and an action plan with owners and a you-said-we-did update. Use for an annual survey, a pulse check or reading results someone else collected. Not for individual performance feedback (use performance-review-cycle) or rituals (use team-rituals-and-recognition).

Reviews staff expense claims and company card spending against the owner's expense policy: reads every receipt, matches it to the claim or card line, checks limits, business purpose, tax reclaim and duplicates, chases missing receipts, and recommends approve, query or reject per line. Use when claims or card statements need checking before reimbursement or month end. Not for coding the bank feed: use categorise-bank-transactions. You recommend; the owner approves and pays.

Designs and builds a finance KPI dashboard as a real spreadsheet: picks 6 to 10 KPIs that fit the business model, defines each precisely with source, owner, target and direction, lays out monthly data, formula-driven KPIs, status tiles and 13-month trend charts, and checks every figure against the ledger. Use when the owner wants to see the numbers that matter at a glance, or a board wants a monthly KPI page. Not for the full monthly pack: use management-accounts-pack.

Builds a driver-based scenario model in a real spreadsheet with a scenario switch (base, upside, downside, plus specific what-ifs such as losing the largest customer or hiring three people), a sensitivity ranking of which drivers matter most, break-even and runway, and a summary the owner can decide from. Use when the owner asks what happens if something changes, whether they can afford a decision, or how bad a bad year could get. Not for the weekly cash position: use thirteen-week-cash-forecast; for the annual plan use budget-build.

Explains the owner's own profit and loss, balance sheet and cash flow statement in plain English: checks the three statements agree with each other, bridges profit to cash, computes the ratios that matter for their business, and answers their actual question with their numbers. Use when the owner asks what their accounts mean, why profit and cash differ, or whether the business is healthy. Not for building the monthly reporting pack: use management-accounts-pack.

Plans a new hire's first five working days hour by hour: arrival or login logistics, setup, manager and buddy time, intro meetings, focused learning, a first small result by day 3 to 5, and a Friday check-in, with calendar invites drafted for approval. Use in the week before a start date. Not for the 90-day arc (use onboarding-plan) or policies (use draft-hr-policy).

Builds or refreshes the ideal customer profile and buyer personas from the owner's own won, lost and churned deals: fit criteria, disqualifiers, buying triggers, personas and an account fit score with tiers. Use when the team needs to decide who to sell to, before building lists or writing outreach. Not for finding the actual companies and contacts (use prospect-list-building) or researching one account (use account-research).

Drafts sales invoices and credit notes that meet the tax rules of the seller's country, match the contract or order, and are numbered, dated and taxed correctly, then leaves them as drafts for the owner to approve and send. Use when work has been delivered and needs billing, an invoice is wrong, a customer needs a refund or discount documented, or recurring invoices need setting up. Not for chasing unpaid invoices: use receivables-chasing.

Sets up and runs leave and absence tracking: a tracker with one row per absence, entitlement and pro-rata accrual formulas, balances, public holidays per location, a monthly absence report, and supportive return-to-work prompts, while keeping health details out of the tracker. Use for holiday balances, sickness patterns or leave reporting. Not for writing the leave policy (use draft-hr-policy). Formal absence cases go to a qualified human.

Produces the monthly management accounts pack from closed books: a one-page summary, P&L for the month and year to date against budget and last year with variances decomposed into causes, balance sheet with movements, cash flow, debtors and creditors, KPIs, and commentary that explains why and what to do. Built as a spreadsheet that ties to the ledger, plus a short memo. Use after month-end close for the owner, board, investors or lender. Not for closing the books: use month-end-close first.

Runs the month-end close for a small business on a five-working-day calendar: cut-off, reconciliations, accruals, prepayments, deferred income, depreciation, payroll and tax control accounts, balance sheet schedules, a flux review of the P&L, and a lock recommendation, with every journal drafted and evidenced. Use when a month has ended and the books need to be right before reporting. Not for the report itself: use management-accounts-pack once the close is done.

Prepares the owner for a commercial negotiation on price and terms: both sides' alternatives and limits, the target and walk-away, a give-get concession plan, packaged offers, answers to common procurement tactics, and the terms that need a lawyer. The owner sets every limit and makes every concession; the agent never agrees anything with the buyer. Use when a deal reaches pricing, procurement or contract talks, or a renewal is being renegotiated. Not for early objections (use objection-handling) or writing the proposal (use proposal-and-quote).

Diagnoses and answers sales objections (price, timing, no need, happy with the current tool, competitor, no authority, risk and trust, missing feature, send me info) by finding the real concern before responding, then drafts an honest, evidence-backed reply and keeps a shared objection library. Use when a prospect or customer pushes back in a call, email or proposal review. Not for trading price and terms in a late-stage negotiation (use negotiation-prep) or a customer asking to cancel (use cancellation-save-offer).

Builds a new hire's onboarding plan from offer acceptance to day 90: pre-boarding tasks with owners and dates, 30/60/90-day outcomes agreed with the manager, a buddy, check-ins, and a short new-hire survey at day 30 and 90. Covers compliance, clarity, culture and connection. Use as soon as an offer is accepted. Not for the hour-by-hour first week (use first-week-schedule) or writing handbook pages (use write-handbook-section).

Sets up and supports manager one-to-ones: cadence, a shared running agenda the employee owns, templates for regular, first, career and skip-level one-to-ones, a question bank, GROW coaching prompts, private and shared note rules, and action tracking between meetings. Use when a manager starts with a new report, wants better one-to-ones, or asks for prep. Not for formal reviews (use performance-review-cycle). Grievances or health disclosures go to HR.

Prepares and checks payroll inputs before anyone is paid: confirms period and roster, collects starters, leavers, pay changes, hours, leave and bonuses, flags every anomaly against last period, reconciles headcount and gross pay, and hands a validated input sheet to the owner or payroll provider. Use before each pay date. Not for staff expense claims: use expense-claim-review. You never run, approve or file payroll or pay anyone.

Designs and runs a performance review cycle: timeline, self, peer, upward and manager review forms, a defined rating scale, manager guidance on evidence-based feedback, calibration sessions with bias checks, delivery conversations and a completion tracker. Use when the company sets up or runs reviews. Not for weekly check-ins (use one-to-one-meetings). Underperformance plans, warnings and dismissals go to a qualified human.

Runs the weekly pipeline review and the period forecast: coverage against target, stage conversion from the owner's own history, deal ageing and slippage, inspection of the deals that matter, commit, best case and pipeline categories based on what buyers actually did, a forecast range, and accuracy tracking over time. Use for the weekly sales meeting, a month or quarter-end forecast, or when the owner asks if the number will land. Not for fixing CRM data (use crm-hygiene) or scoring a single deal in depth (use discovery-qualification).

Turns discovery notes into a proposal and quote the buyer can say yes to: their problem in their words, outcomes and success measures, scope with what is out, options, a checked price table, terms, assumptions and next steps, plus a go or no-go call on RFPs. Prices, discounts, terms and promises are left for the owner to confirm; nothing is sent without approval. Use after discovery when the buyer has asked for a proposal or quote. Not for qualifying the deal (use discovery-qualification) or negotiating terms (use negotiation-prep).

Builds a researched, scored list of target accounts and the right contacts inside them from public sources and the owner's CRM, with a reason to contact, a source and a lawful basis on every row, deduplicated against customers, open deals and opt-outs. Never buys or scrapes lists. Use when the team needs who to contact next. Not for deep research on one account (use account-research) or writing the messages (use cold-outreach-sequence).

Builds a quarterly business review deck for a customer: results against the goals they set, value delivered in their numbers, adoption, support summary, what is not working and the fix, relevant roadmap, recommendations, next quarter's goals and mutual asks, plus the pre-meeting check with the champion and the follow-up. Use before a QBR, executive review or annual review with an existing customer. Not for the internal account plan (use account-plan) or the renewal deal itself (use renewal-and-expansion).

Works the aged receivables: checks each overdue invoice is really unpaid and undisputed, scores each customer by payment history, drafts one tone-matched reminder per customer on a fixed cadence, tracks promises to pay, and flags accounts for the owner to escalate. Use when the owner asks who owes them money, overdue invoices are building up, or cash is tight. Not for raising or correcting invoices: use invoices-and-credit-notes.

Runs renewals and expansion deals with existing customers: a renewal timeline worked back from the contract end and notice dates, the value review, the renewal quote and price-change handling, what to do when a renewal stalls, renewal forecasting with gross and net revenue retention, and spotting and qualifying expansion. Prices, discounts and terms are the owner's call; every customer message is drafted for approval. Not for the strategic account plan (use account-plan) or a customer asking to cancel (use cancellation-save-offer).

Prepares the owner or a teammate for an upcoming sales call: a one-page prep sheet with the call's single objective, attendee research, context from past conversations, an agenda the buyer agrees to, questions to ask, likely objections, proof to use and the next step to ask for, plus cold-call and voicemail scripts. Use before any cold call, discovery call, demo, follow-up or executive meeting. Not for scoring the deal afterwards (use discovery-qualification) or researching an account in depth (use account-research).

Drafts follow-ups for live sales conversations: the recap after a meeting, the nudge after a proposal, mutual action plan updates, re-engaging a deal that has gone quiet, multi-threading when the champion goes silent, and the respectful close. Each draft adds something new and is shown to the owner before it is sent unless the owner has said to send. Use when someone has already engaged. Not for first contact with cold prospects (use cold-outreach-sequence) or price and terms talks (use negotiation-prep).

Screens CVs and applications against criteria fixed before reading them: redacts details that invite bias, scores each criterion with quoted evidence, flags borderline cases for a human, and drafts rejection notes for the owner to approve. Use when applications arrive for a role. A human makes every final decision. Not for finding people who have not applied (use source-candidates) or for interview questions (use structured-interview-kit).

Runs a blameless post-incident review once a security incident or near miss is contained: rebuilds the timeline from evidence, measures detection and response times, finds the contributing causes, and turns them into tracked actions with owners and dates. Also prepares a shorter external version for customers, insurers or regulators. Not for handling an incident that is still active (use security-incident-triage).

Finds candidates for an open role from public professional profiles only (search engines, professional networks, GitHub, portfolios, talks), keeps a minimal private talent list, and drafts personal outreach for the owner to approve. Use when a role needs more or better applicants than the ad brings. Not for judging applicants who already applied (use screen-cvs-fairly) or writing the ad itself (use write-job-description).

Builds a structured interview kit for a role: competencies mapped to stages, the same questions for every candidate, planned follow-up probes, anchored 1 to 5 scorecards, an interviewer brief with questions never to ask, and a debrief that collects scores before discussion. Use before interviews start or when a process feels like gut feel. Not for screening CVs (use screen-cvs-fairly) or drafting the offer (use draft-offer-letter).

Prepares a supplier payment run for the owner to approve and release: checks each bill is genuine, approved, not a duplicate and not disputed, verifies bank details, nets credit notes, weighs early-payment discounts, checks the cash position first, and produces a run sheet and remittance drafts. Use on payment day, when bills pile up, or when the owner asks what they owe and what to pay this week. Not for customer invoices: use invoices-and-credit-notes. You never make the payment.

Builds and maintains a rolling 13-week direct cash flow forecast in a real spreadsheet: opening cash from the reconciled bank balance, receipts by customer from actual payment behaviour, payments by due date, the weekly low point against a minimum buffer, and a weekly forecast-versus-actual review. Use when the owner asks whether they can make payroll, how long cash lasts, or what to do about a shortfall. Not for an annual budget or monthly P&L plan: use budget-build or financial-scenario-model.

Works out what one customer, order or project is really worth: fully loaded acquisition cost, contribution margin after every variable cost, lifetime value from real cohort retention, payback period and LTV to CAC, by channel and segment, in a spreadsheet that ties back to the P&L. Use when the owner asks if growth is profitable, which channel or product to push, what they can afford to spend to win a customer, or how pricing changes the economics. Not for the company-wide budget: use budget-build.

Runs a periodic user access review across the owner's systems: exports every account and permission, matches them to the people roster, flags leavers, dormant, shared and over-privileged accounts, missing MFA, stale keys and risky app grants, then records each access owner's keep or remove decision as audit evidence. Use for quarterly reviews, after someone leaves, or for SOC 2 or ISO 27001 evidence. Not for hardening system settings (use server-and-cloud-hardening).

Handles security questionnaires in both directions: answers incoming ones from customers and prospects (SIG, CAIQ or custom spreadsheets) truthfully from evidence and a reusable answer bank, and assesses vendors the owner wants to use, scaled to the data and access they will have. Use when a security questionnaire arrives or before signing a supplier that will touch company or customer data. Not for a full framework gap analysis (use compliance-readiness-assessment).

Checks a web application the owner owns, with written authorisation in this conversation, against the OWASP Top 10 and ASVS: HTTPS and headers, cookies, login and session handling, and access control between the owner's own test accounts, using non-destructive checks only. Refuses apps the owner does not own. Use for a pre-launch check or 'is our web app secure?'. Not for servers and open ports (use authorised-vulnerability-assessment) or reading source code (use secure-code-review).

Finds out why deals are won and lost: pulls closed deals from the CRM, codes primary and contributing reasons from notes and emails, interviews buyers (requests drafted for the owner's approval), compares won and lost deals by segment, source, competitor, size and qualification gaps, and turns patterns into ranked recommendations with owners. Use quarterly, after a run of losses, or when a competitor keeps appearing. Not for inspecting open deals (use pipeline-review-and-forecast) or customer churn (use churn-risk-signals).

Writes or rewrites an employee handbook section (how we work, time off, expenses, equipment, benefits, communication norms) in plain, scannable English, with an owner, a review date, links to the binding policy, and every legal statement sourced and marked for qualified review. Use when a new hire or teammate keeps asking the same question or a page is out of date. Not for the formal, binding policy itself (use draft-hr-policy).

Runs a short intake with the hiring manager and turns it into an outcome-based job description and job ad: clear must-haves, a pay range where the law or the owner wants one, inclusive wording and a lawful equal opportunity line. Use when a new role opens or an old ad gets no good applicants. Not for scoring applicants (use screen-cvs-fairly) or writing interview questions (use structured-interview-kit).

Reviews the owner's logs (servers, web, cloud audit, email and identity) and hunts for signs of compromise with written hypotheses and baselines: logins after many failures, logins from new places, new admins and keys, persistence, logging switched off, rare outbound traffic and odd data transfers, mapped to MITRE ATT&CK. Read-only. Use for a routine log review or 'has anyone got in?'. Not for tuning noisy alerts (use security-alert-tuning) or running a confirmed incident (use security-incident-triage).

Cuts security alert noise without losing real detections: measures each rule's volume and precision, finds why false positives fire, tunes logic, scope, thresholds and allowlists (each with an owner and expiry), proves the tuned rule still catches a true positive, and documents every detection. Use when alerts are too noisy, ignored, or missing things. Not for searching logs for compromise (use log-threat-hunting) or handling an alert that is a real incident (use security-incident-triage).

Writes and maintains the owner's information security policies (top-level security policy, acceptable use, access control, incident response, backup and continuity, vendor management, data classification and retention, secure development) in plain language that matches what the company really does, mapped to SOC 2, ISO 27001 or GDPR where needed. Use when a customer, auditor or new team needs policies. Not for measuring gaps against a framework (use compliance-readiness-assessment).

Builds a realistic study plan for a course, exam, qualification or new skill: breaks the syllabus into topics with weights, rates the learner's confidence, fits sessions into the hours they really have, schedules spaced reviews and mixed practice, adds buffers and weekly check-ins, and puts it in their calendar. Use when a learner says make me a study plan, I have X weeks until Y, or how should I learn Z. Not for the final weeks of exam cramming (use exam-revision-sprint) or explaining a topic (use explain-at-learner-level).

Handles a customer who asks to cancel or downgrade: finds the real reason, matches one fitting, owner-approved option (fix the problem, right-size the plan, pause, training, a time-limited discount) or none, makes cancelling easy and lawful with no obstruction, and confirms dates, data and final charges. Replies are drafted for the owner's approval and every offer stays within the owner's rules. Use when a cancellation or downgrade request arrives, or to design the save process. Not for spotting at-risk customers early (use churn-risk-signals).

Finds the customers likely to leave before they say so: combines usage, support, billing, relationship and contract signals into a documented health score, flags at-risk accounts with dated evidence, back-tests the score against past churn, and hands the owner a ranked watchlist with the reason and a suggested action for each. The analyst does not contact customers. Use for a weekly risk review, before renewals, or after churn rises. Not for handling a customer who has asked to cancel (use cancellation-save-offer) or survey analysis (use nps-csat-analysis).

Runs the owner's customer community (Discord, Slack, a forum or similar): purpose, guidelines and channels, every question answered in the owner's tone, complaints moved to private support, fair and logged moderation, welcoming new members, health reports and incident handling. Announcements and anything that speaks for the company are approved by the owner first. Use for daily upkeep, setting a community up, or a moderation problem. Not for one-to-one tickets (use support-ticket-reply) or routing feedback to product (use product-feedback-loop).

Researches and compares options before the owner books or buys: hotels, flights, rentals, tradespeople, appliances, insurance, broadband, services. Sets must-haves and deal-breakers, compares true total cost and terms across direct and comparison sites, reads reviews for patterns and fakes, checks the seller is legitimate, scores options and recommends one, then books only after explicit approval and tracks the cancellation deadline. Use for which should I book or buy, find me the best X, compare these. Not for a whole trip itinerary (use plan-a-trip).

Builds and keeps a register of every supplier contract and subscription with end dates, auto-renewal terms, notice deadlines, value and price-increase terms, finds untracked subscriptions in bank and card statements, sets reminders against the notice deadline rather than the end date, and prepares a renew, renegotiate, replace or cancel brief for each renewal. Use when the owner wants control of contracts, a renewal or price rise lands, or spend needs cutting. Not for choosing a new supplier: use vendor-selection.

Finds meeting times that work across calendars, time zones, working hours and public holidays: reads free and busy time, computes overlap with real time-zone rules including daylight saving changes, ranks options by how far each person is pushed outside their day, rotates the pain for recurring meetings, and proposes three options shown in every attendee's local time. Use when the owner or team needs a meeting across locations, or a recurring slot that is fair. Not for planning people's workload across weeks: use resource-capacity-planning.

Plans and runs onboarding for a new customer: takes the handover from sales, agrees the first measurable win with the customer, builds a shared 30-60-90 day success plan with commitments on both sides, schedules check-ins, watches for stalls and acts on them, and hands over to account management when the customer has graduated. Every message to the customer is drafted for the owner's approval. Use when a customer signs or starts a trial. Not for rescuing an at-risk existing customer (use churn-risk-signals) or written help articles (use help-centre-article).

Plans and runs decluttering and home projects: room or category declutter sessions with clear keep, donate, sell and dispose rules, safe disposal and selling, then home improvement and repair projects with scope, a DIY-or-professional decision, permits to check, comparable quotes, a budget with contingency, a materials list and a sequenced schedule. Use when the owner says help me declutter, we're moving, plan the bathroom redo or get quotes for X. Not for routine bills and paperwork (use household-admin-and-bills).

Runs the last 1 to 8 weeks before an exam: a specification checklist rated by confidence, timed past papers marked against the official scheme, an error log that sorts mistakes by cause, targeted retrieval and mixed practice, exam technique by question type, and a calm final-week and exam-day plan. Use when a learner says my exam is in X weeks, help me revise, or I keep losing marks. Not for long-range planning from the start of a course (use build-study-plan) or one-off marking (use mark-work-with-feedback).

Teaches a topic one to one at the learner's actual level: probes what they already know, explains with concrete examples, worked examples and diagrams, checks understanding by having them explain and apply it, and dislodges misconceptions with counter-examples. Use when a learner says I don't get X, explain Y simply, or help me understand this homework. Not for planning weeks of study (use build-study-plan) or writing a quiz (use make-practice-questions).

Runs a household's shared calendar: one source of truth for every family member, school terms and activities imported, a weekly logistics grid of who drives whom and when, conflict and clash detection, deadlines for forms, payments and kit, and a short weekly family briefing. Use when the owner says sort out the family calendar, who's doing pickup, what's on for the kids this week, or we keep double-booking. Not for one person's work week (use plan-the-week) or trip itineraries (use plan-a-trip).

Keeps track of the household's important dates and plans for them: thoughtful gift shortlists from what is known about each person, ordered in time to arrive, with card messages; and small to medium events such as birthday parties, dinners, family gatherings and celebrations, with a brief, budget, guest and RSVP tracker, dietary needs, supplier comparison, a countdown and a run-of-show for the day. Use for gift ideas, birthday reminders or planning a party. Not for full trips (use plan-a-trip).

Writes and maintains help-centre articles and FAQs from real support tickets: picks topics by ticket volume, chooses the right article type (task, troubleshooting, reference, explanation or FAQ), writes titles in the customer's own search words, verifies every step in the product, adds redacted screenshots, and tracks whether the article cut tickets. Published only after the owner approves. Use when customers keep asking something the help centre does not answer well. Not for internal saved replies (use support-macros) or answering a single ticket (use support-ticket-reply).

Organises household paperwork and money admin: finds every recurring bill and subscription from statements and emails, builds a bills calendar with due dates, amounts and renewal dates, flags price rises and unused subscriptions, tracks deadlines such as renewals, tax and licences, sets reminders, and sets up a simple filing system. Use when the owner says sort out our bills, what are we paying for, when is X due, or I keep missing deadlines. Not for comparing replacement deals in depth (use compare-and-book-research). It never pays, cancels or files anything itself.

Writes practice questions, quizzes and mock tests that measure what was taught: questions aligned to learning objectives and the exam's style, a planned mix of difficulty and thinking levels, multiple-choice items with distractors built from real misconceptions, verified answer keys with explanations and mark schemes, and output as a printable quiz, a flashcard file or an interactive quiz. Use when asked for a quiz, practice questions, a mock test or flashcards. Not for marking a learner's answers (use mark-work-with-feedback).

Marks a learner's work (answers, essays, problem sets, practice papers, projects) against the mark scheme, rubric or level descriptors, with evidence for every mark, then writes specific feedback the learner can act on: what was done well, the two or three most important targets, and a follow-up task. Also produces class summaries of common errors for teachers. Use when asked to mark this, grade my essay, check my answers or give feedback. Not for writing the questions (use make-practice-questions).

Plans a week of meals for a household's tastes, allergies, schedule, equipment and budget, reuses ingredients to cut waste, scales recipes, checks the pantry, and produces a shopping list grouped by aisle with quantities and an estimated cost, price-checking only the items that matter. Use when the owner says plan our meals, what's for dinner this week, make a shopping list, or we waste too much food. Not for clinical diets, which need a dietitian, or for event catering (use gift-and-event-planning).

Analyses customer survey results (NPS, CSAT and customer effort): calculates scores correctly with margins of error, segments by plan, tenure and other groups with minimum sample sizes, codes open-text comments into themes, finds what separates promoters from detractors, and turns it into a short report with actions and a close-the-loop list. Follow-up messages to respondents are drafted for the owner's approval. Use after a survey closes, for a monthly CSAT review, or when a score moves. Not for finding at-risk accounts from usage (use churn-risk-signals).

Investigates an operational failure or recurring problem without blame: contains the impact first, defines the problem precisely (what, where, when, how much, and what it is not), rebuilds the timeline from records, uses Pareto, fishbone and evidence-checked 5 whys to find causes and contributing factors, verifies them, and sets corrective and preventive actions with an effectiveness check. Use after a missed delivery, a costly error, a complaint spike or a process breakdown. Not for reviewing a whole project: use project-retrospective.

Builds an operations dashboard that separates real change from noise: a metric tree from the outcome the owner cares about to the drivers the team controls, precise definitions, weekly data pulled from the systems of record, process behaviour (XmR) charts with natural limits and signal rules, and a short weekly read-out of what changed and what to do. Use when the owner wants to see how operations are running, or a metric moved and they want to know if it matters. Not for finance KPIs: use finance-kpi-dashboard; for explaining a specific failure use operational-root-cause-analysis.

Plans a trip end to end: a brief of who, when, budget and pace, entry and health requirements checked on official sources, a realistic day-by-day itinerary grouped by area with travel times, where to stay and how to get around, a budget, a booking timeline, a documents folder and a packing list. Use when the owner says plan a trip to X, we're going away in May, or build an itinerary. Not for comparing one specific purchase or booking in depth (use compare-and-book-research) or for weekly schedules (use plan-the-week).

Runs the owner's weekly review and plans the week ahead: gathers commitments from calendar, email and task lists, picks the three outcomes that matter, time-blocks focus work, errands and personal time with realistic buffers, flags conflicts and deadlines, then sends a short daily brief. Use when the owner says plan my week, what's on this week, I'm overwhelmed, or help me prioritise. Not for coordinating several family members' calendars (use family-calendar-coordination) or trip itineraries (use plan-a-trip).

Improves a slow, costly or error-prone process with evidence: defines the problem with baseline numbers, finds the bottleneck and the waste, separates value-adding work from waiting and rework, designs a future state (eliminate, combine, simplify, then automate), pilots it, proves the change with before-and-after data and locks it in with a control plan and a business case. Use when the owner says something takes too long, costs too much or keeps going wrong. Not for documenting a process as it is: use process-mapping-and-sops.

Captures how a process really runs and writes it down so anyone can follow it: SIPOC to set the boundaries, a swimlane map with handoffs, decisions and exceptions, per-step time and quality measures, then a standard operating procedure with numbered imperative steps, checks, exceptions and an owner, tested by someone following it cold. Use when work lives in one person's head, a role is being handed over, errors come from inconsistency, or an audit needs documented procedures. Not for redesigning a process: use process-improvement.

Runs the loop from customer feedback to product and back: captures requests, bugs and complaints from tickets, calls, community, surveys and cancellations with the customer's own words and evidence, rewrites them as the underlying problem, deduplicates them, weighs them by accounts and revenue, reports them to product, and tells customers when something ships or will not be done. Messages to customers are drafted for the owner's approval; no roadmap promises. Use weekly or monthly, or when a request keeps coming up. Not for running the community itself (use community-management).

Turns a goal into a project plan people can run: a one-page charter with scope and success measures, a deliverable-based work breakdown, three-point estimates from the people doing the work, dependencies, the critical path computed in Python, a dated Gantt in a spreadsheet, RACI, milestones and a baseline with change control. Use when the owner or a teammate starts a project, asks when something will be done, or a plan needs rebuilding. Not for weekly progress updates: use project-status-report.

Runs a blameless retrospective on a finished project or phase: builds a factual timeline and plan-versus-actual numbers, gathers views from everyone involved separately, finds the few themes that matter and their causes, and agrees three to five owned actions that are followed up, with lessons saved for next time. Use when a project, launch or quarter has ended, or a phase went notably well or badly. Not for a single operational failure or incident: use operational-root-cause-analysis.

Builds and runs a project risk register: finds risks with a pre-mortem, assumption review and category prompts, writes each as cause, event and effect, scores probability and impact on defined 1 to 5 scales, assigns an owner, response, trigger and contingency, tracks residual risk and moves risks that happen to an issue log, with a heat map spreadsheet. Use at project start, before a big decision, or weekly during delivery. Not for investigating a failure that already happened: use operational-root-cause-analysis.

Writes an honest weekly or monthly project status report from evidence: overall and per-dimension RAG status against defined rules, milestones baseline versus forecast, progress since last report, earned value where cost data exists, top risks and issues, and the decisions needed with dates, in an executive and a team version. Use when the owner or a sponsor needs to know where a project stands. Not for building the plan itself: use project-plan-and-timeline; for a project that has ended use project-retrospective.

Takes a purchase request from need to approved purchase order: checks the need and existing stock or contracts, checks budget, gets the right number of like-for-like quotes for the amount, routes it through the owner's approval limits, onboards new suppliers safely, drafts the PO, and records delivery for three-way match. Use when someone needs to buy goods or services, or the owner wants spending controlled. Not for choosing a long-term supplier: use vendor-selection. You never place orders, accept quotes or pay.

Compares the work the team has committed to with the hours people actually have, week by week: real capacity after leave, public holidays and standing duties, demand from projects and day-to-day work, a utilisation heat map in a spreadsheet, the overloaded weeks and people, and options to resequence, reassign, buy in help or cut scope. Use when the owner asks if the team can take on more, a deadline looks crowded, or people seem overloaded. Not for booking a single meeting: use cross-timezone-scheduling.

Turns a PDF, textbook chapter, article, slide deck, lecture recording or video into a study pack: section-by-section notes with page or timestamp references, a layered summary, glossary, concept map, worked examples, retrieval questions and importable flashcards, covering every section and marking what came from the source versus your additions. Use when a learner says help me learn this chapter, summarise this lecture or make notes from this video. Not for a full quiz with mark scheme (use make-practice-questions).

Triages a support inbox or helpdesk queue: files what needs no answer, spots spam and phishing, categorises and prioritises the rest by impact and urgency, routes each item to the right person, drafts routine replies from approved macros, detects incidents from clusters of similar reports, parks anything needing the owner's judgement with a one-line reason, and tracks response targets. Nothing is sent until the owner has read it. Use for a daily or hourly inbox sweep. Not for writing a full answer to one ticket (use support-ticket-reply).

Builds and maintains the saved replies (macros) support uses for recurring questions: finds the top contact reasons from ticket data, writes macros with placeholders and a required personal line, adds the internal actions (tags, status), sets owners and review dates, and retires unused or outdated ones. Macros speak for the company, so the owner approves them before they go live. Use when the same questions keep coming in or the macro library is stale. Not for answering one ticket (use support-ticket-reply) or public help articles (use help-centre-article).

Answers a customer support ticket properly: reads the whole thread and account history, works out the real question, checks the answer against the help centre, product and past tickets, writes a clear, kind reply in the owner's voice, and escalates with a full handoff note when it cannot be answered safely. Refunds, credits, dates and promises go to the owner; replies are drafted for approval unless the owner has said to send. Use for any single ticket, chat or email. Not for sorting a whole inbox (use support-inbox-triage) or saved replies (use support-macros).

Runs a fair, evidence-based vendor or supplier selection: written requirements with must-haves, a long list cut to a short list, a request for quotes, scripted demos, weighted scoring with weights fixed before scoring, a three-year total cost of ownership, due diligence and reference checks, and a recommendation with a runner-up and the contract points to negotiate. Use when the business needs a new supplier, tool or service provider, or wants to test the market before a renewal. Not for tracking existing contracts: use contract-renewal-tracking.