What it is made of, and how it runs.

Synthesises daily status files, transcripts, issue logs and vendor scorecards into an audit-grade narrative report. Every claim sourced; vendor-attributed claims gated.

When
The recurring reporting cycle; the anchor the deck and infographic build from.

Builds the governance PowerPoint from the status report plus the prior deck, with brand binding and a full per-slide QA battery.

When
Every governance cycle.

A self-contained interactive HTML view with three audience depths and drill-down to atomic claims with source citation.

When
The live operational view; pre-meeting refreshes.

Generates the Claude Design prompt that produces the same three-view prototype via the chat-plus-canvas loop.

When
Iterating the visual in Claude Design rather than building directly.

A reasoning engine: hand it any artefact and it calibrates is this normal, is this a real blocker, what would an experienced hand see. Backed by an evidence-graded research base.

When
A second read on programme health, especially under pressure.

A structured pre-mortem: assumes failure, works backwards to failure modes, leading indicators and the verification that would have caught each.

When
Before gate decisions; periodic stress tests.

The earlier live operational dashboard, retained; superseded in daily use by the infographic.

When
Legacy; scheduled refresh hooks remain.

A clear brief: the problem, success criteria, and what is in and out of scope.

When
Turning a fuzzy idea into a brief, before any building.

The shape of the solution and the key trade-offs, before anyone builds.

When
Deciding how the whole thing fits together.

Your information structured so it holds together, with quality rules.

When
When a solution depends on data.

A sound build plan: what to use, where the risks and shortcuts are.

When
Before building the working thing.

A screen, form or page people can actually find their way around.

When
When a solution has something people interact with.

How it could break, the edge cases, and proof it does what the brief promised.

When
Before it goes live.

What the data actually says, with the analysis and charts kept truthful.

When
When the question is what does this data say.

Context-corrects raw meeting transcripts into clean merged markdown, with a per-file audit log and a versioned lexicon applied.

When
Cleaning a raw transcript before minutes.

Manages a versioned lexicon of canonical names and terms with a staged propose-review-commit workflow.

When
Keeping terminology canonical across documents.

Chair-ready minutes from a corrected transcript plus attendance. An intake readback gate, a Claims Verification Table sourcing every claim, then the fact-checker as a separate verification layer; high-severity findings halt the ship.

When
Producing defensible minutes.

PowerPoint to a board-scrutiny bar: native primitives only, a 13-gate programmatic battery, and a per-slide visual walkthrough. Two-gate done.

When
Decks that must withstand scrutiny.

Files any folder into the workspace taxonomy by sampling contents, proposing a reviewable ledger first; commits only approved, reversible moves.

When
Tidying a folder safely.

The installer and reconciler for the whole estate: doctor, plan, heal, with backup, snapshot, validate and rollback. Appends a dated changelog line to every skill it edits.

When
Installing, syncing or repairing the estate.

Packages skills into a self-contained payload (verbatim files, per-file hashes) plus an executor prompt that rebuilds them on the other machine, hash-verified, with backup.

When
Moving skills between the two laptops.

Turns a rough idea into a signed-off brief, then production-quality images or video, picking the right model and critiquing every output. Every spend pre-costed and gated.

When
Any image or video need.

Expert web engineer for sites hosted on a personal domain: publish a build, push an update, configure hosting (SSL, DNS, email, databases), or advise before a build. Four modes, each interview-first, reference files loaded only when needed.

When
Publishing or maintaining a hosted website.

Claude.ai chat prompts: a lean and a full version, writing rules applied, grounded in the published technique stack.

When
Building a chat prompt worth reusing.

Cowork agentic briefs: depth-first, files in and out, persistent logs, skill routing, standing guardrails and a verification step.

When
Setting an agent on a real task.

Claude Design prompts: researches current capabilities every run, maximises interactivity, prepares the upload-ready files and says what to upload.

When
Creating a visual artefact in Claude Design.

Audit-grade Microsoft 365 Researcher prompts that defend against that agent's known failure modes with source inventories and evidence ledgers.

When
A deep cross-source mailbox or document audit.

Sets up and maintains whole projects: scaffolds instructions, CLAUDE.md, logs, tasks, decisions and references; nothing built before a confirmed readback.

When
Standing up or maintaining a project.

Works a hard call through frame, outside view, three-plus options, a bias check and the strongest opposing case, written to a decision record. Strengthens reasoning, does not confirm it.

When
A consequential choice, work or personal.

An adversarial verifier: extracts every checkable claim and verifies against primary sources, grading with severity. Default posture HALT; never fabricates a source.

When
Before anything is sent or published.

Turns a half-formed idea into a build-ready spec, pushing on edge cases and handoffs, then recommends the right builder. Challenges the premise first.

When
When an idea is not yet a spec.