Skillsdbhq.uk (external site)

skills.dbhq.uk

Sixteen skills for your coding agent

Free, open-source skills that extend Claude Code and Codex with capabilities drawn from real delivery work - each installs from the DBHQ marketplace with a single command, no sign-up.

Two commands, then you have it

  1. Add the DBHQ marketplace, once

    /plugin marketplace add dbhq-uk/marketplace
  2. Install any skill by name

    /plugin install <name>@dbhq

Both commands work in Claude Code and in Codex. The marketplace is a public repository you can read first: dbhq-uk/marketplace (external site)

The sixteen

  • outlook

    Claude Code + CodexLive

    Microsoft 365 mail and calendar, from the terminal

    Reads and sends Microsoft 365 mail, runs a calendar, and handles drafts, attachments and availability over the Graph API with no MCP server. Replies keep every recipient by default. A second skill turns a PST export into markdown with a SHA256 chain of custody.

    /plugin install outlook@dbhq

  • trello

    Claude Code + CodexLive

    Your boards, run from your agent

    Trello skills in one install, over the REST API: boards, lists and cards end to end, a status digest for a board, a radar for what is slipping across all of them, and a shopping list sorted into supermarket aisle order.

    /plugin install trello@dbhq

  • legwork

    Claude Code + CodexLive

    Research that settles a decision, and says when it cannot

    Multi-source research where every claim carries how well it is supported, a source is judged by the claim it backs rather than by its domain, and corroboration counts only across sources that could have disagreed. A run that settles nothing says so.

    /plugin install legwork@dbhq

  • dovetail

    Claude Code + CodexLive

    Checks whether your repository still agrees with itself

    Seventeen deterministic checks over a repository - dead links, moved heading anchors, orphaned files, duplicated content, documented flags a script does not accept, manifests that disagree about a version - with no model, no network and no packages, and every finding marked proven or heuristic. Model reviewers add what Python cannot settle, such as contradictions between documents.

    /plugin install dovetail@dbhq

  • verve

    Claude Code + CodexLive

    Strips AI tells from prose and puts a voice back

    Strips the tells that mark writing as machine-made and puts a voice back, in British or American English, pitched at a named reader. Four tone presets, three strengths, a two-way politeness check, and meaning as a veto: no figure, name or claim changes.

    /plugin install verve@dbhq

  • vela

    Claude Code + CodexLive

    Compiler-exact code search, in any language you index

    Definitions, references, direct callers and implementations in about a second. Roslyn-exact for C#, Visual Basic, Razor and Blazor, whose compiler-resolved references it saves to an index you query with no server running, and any other language through its SCIP index, in the same database.

    /plugin install vela@dbhq

  • garmin

    Claude Code + CodexLive

    Your Garmin data, answered in the terminal

    Body Battery, HRV, resting heart rate, stress, sleep stages and score, activities, VO2 max, training load and readiness, from Garmin Connect. Answered live, or written down as a day's markdown snapshot and a week's rollup you keep. The password is never stored, and nothing but Garmin is ever called.

    /plugin install garmin@dbhq

  • imager

    Claude Code + CodexLive

    Images from OpenAI, costed before it spends

    Generates and edits images with OpenAI's GPT Image models - 27 style presets, platform sizing, transparent backgrounds, masked edits and carousels held together by a reference image. It prices the run before it spends, and a draft costs a fraction of a final, so you iterate on drafts and pay for the one you approved.

    /plugin install imager@dbhq

  • gitview

    Claude Code + CodexLive

    Which branches are finished, and which only look like it

    One table per repository: worktree, uncommitted changes, branch, open pull request, ahead, behind, unpushed, and whether the branch can go. It decides the last column by test-merging with git merge-tree, because a squash-merged branch stays ahead of the trunk forever and every obvious check gets it wrong. Work that exists nowhere but this clone is named first.

    /plugin install gitview@dbhq

  • atlassian

    Claude Code + CodexLive

    It edits a real page without losing what it does not understand

    Raises, reads and comments on Jira issues and moves one at a time through its workflow, after checking the project key, the issue type and the required fields against your site. Searches, reads, creates and updates Confluence pages, and publishes a markdown file to a page. REST API and an API token, no MCP server. To change one block it splices your fragment into the live page, so the macros, colours and inline comments around it never pass through the converter.

    /plugin install atlassian@dbhq

  • pennyblack

    Claude Code + CodexLive

    A physical letter, posted from the terminal

    Puts a PDF in the post: printed in the UK, delivered by Royal Mail, with Signed For, Tracked 24 or 48 and Special Delivery, and the tracking number recorded once Royal Mail issues it. Drafting and sending are separate commands and every draft is a test draft until you say otherwise.

    /plugin install pennyblack@dbhq

  • buildwork

    Claude Code, or Codex with PaseoLive

    Your open issues, run as parallel agents

    Puts one agent on each issue you pick, in its own git worktree with its own pull request, then runs mechanical quality control and git merge-tree trial merges over the results and proposes an order to merge them in, with a reason per position. It refuses to fan out when the issues are really one piece of work.

    /plugin install buildwork@dbhq

  • deskwork

    Claude Code + CodexPreview

    What an agent noticed, tracked as real work

    Turns what an agent noticed into a tracked GitHub issue, keeps the dependency edges between issues honest, and writes a roadmap into the repository carrying the date, the issue count behind it and a line of reasoning per call - so the ordering can be argued with.

    /plugin install deskwork@dbhq

  • groupwork

    Claude Code + CodexLive

    A second agent on the work, and a result you can cite

    Four blind patterns for a second agent: red team, second opinion, verify and panel. None is told your conclusion, and that is enforced in code rather than promised in prose. Red team starts in an empty directory rather than your repository, and every run keeps the exact brief it was sent, with its hash in the citation.

    /plugin install groupwork@dbhq

  • headwork

    Claude Code + CodexLive

    One decision at a time, with a recommendation

    Puts the decision blocking you as one question, in plain English, with every option carrying its own justification, one named as the recommendation, and a clause saying what would overturn it. It asks nothing it could have answered by reading the repository first, nothing too cheap to be worth your attention, and one thing at a time.

    /plugin install headwork@dbhq

  • heliograph

    Claude Code + CodexLive

    Debug a machine you cannot log into

    Publish a step, an operator runs it on a machine you cannot log into, and the whole run comes back as a log with every line stamped in UTC - so a hang reads as a gap rather than as slow progress. The station is plain text the operator can read before running it, and nothing holds a connection open.

    /plugin install heliograph@dbhq

What is a skill?

A package of instructions and tooling that teaches a coding agent - Claude Code or Codex - to do something it could not do before. The agent gains a real capability, calls real APIs, and works in your terminal alongside your code. Each of these started as something I needed on an engagement, got hardened until it was reliable enough to depend on, and then got published. If I do not use it myself, it is not here.

Why are they free?

Because they are the most honest demonstration I have. You can read every line, install it in one command and judge the engineering yourself, before you ever talk to me. It is the same judgement you would be buying on an engagement, in a smaller box.

Every page states what its skill will not do

That block is the credibility mechanism this whole site rests on. legwork tells you how well a finding is supported and will say it settled nothing. dovetail refuses to pick a side, because what disagrees is decidable and which side is right usually is not. verve treats meaning as a veto and is not a detector-evasion tool. vela says so when a project fails to load rather than returning a shorter answer. A tool that never names its ceiling is one you cannot calibrate, and a limit you find out about yourself, later, costs more than one you were told.