Reusable workflows for outcome-aligned agents

Use when a code repository needs a README a newcomer can rely on to understand, install, and run the project, and you want it to stay accurate to the actual code; works for any language and handles monorepos.

OutcomeA newcomer can understand, install, and run the project from its README on the first attempt, and the README's commands and links stay accurate to the code.

Use when you can describe a flow, architecture, data model, state machine, class layout, or topic tree in one or two sentences and want a clean, shareable diagram back without learning Mermaid syntax or opening a separate diagramming tool. You get a diagram source you can paste anywhere plus a rendered image you can drop into docs, slides, or chat.

OutcomeA shareable diagram of what you described, ready to drop into documentation, a slide, or a chat thread.

Use when starting a seed or Series A fundraise and the deck structure has not been decided yet, or when a draft feels disorganized and a sanity check against the way investors actually read decks would help. Produces a slide-by-slide outline so the next step is writing slide content, not arguing about which slides to have.

OutcomeFounders walk away with a ready-to-build investor pitch deck outline ordered the way investors actually read decks, cutting deck-structuring from a week of debate to a single review.

Use when the user asks for a slide deck, talk deck, or presentation from a structured outline or a rough description, and wants a real artifact (browser-openable HTML, or a PowerPoint file) rather than just an outline.

OutcomeTurn a slide outline or a rough natural-language description of a talk into a presentable slide deck the user can open and deliver immediately.

Use when reviewing a pull request, branch, or set of code changes for correctness, security, performance, and style issues. Resolves a PR URL, PR number, branch name, or natural-language description into the actual diff, then produces an actionable in-chat review with severity-tagged comments grouped by category, plus concrete unified-diff patches for blockers.

OutcomeWalk a PR or branch and deliver an actionable, in-chat code review grouped by category (correctness, security, performance, style) with severity tags and concrete patches for blockers.

Use when you need real unit tests for a function, class, or module and want them generated, verified, and ready to merge. Describe the target in plain language; the workflow handles discovery and writing, then confirms the tests pass, lift coverage, and lint cleanly before handoff.

OutcomeUnit tests for the named target land in the project, exercise at least one happy path and at least one error path, raise the target file's line coverage, and lint clean.

Use when you want to drive a feature implementation through a strict test-driven cycle of failing-test, minimal-pass, and refactor commits on the current branch.

OutcomeThe feature described in the spec is implemented and covered by a passing test that was written before the implementation, with the work delivered in clean red-green-refactor order on the current branch.

Use when drafting a markdown changelog entry from a git commit range, with user-facing language up top and a smaller technical details section beneath.

OutcomeProduce a markdown changelog entry for a git commit range, written for users at the top with a smaller technical details section beneath.

Use when a user hands over a fillable PDF along with partial or natural-language information about what should go in it. The agent reads the form, asks the user grouped questions for any missing values, fills the PDF, and confirms the result is correct before handing it back.

OutcomeA filled PDF the user can submit, with every requested value verified to read back exactly as written.

Use when the user asks a natural-language data question against a database in the current directory and wants a reviewable SQL query back. The agent discovers the schema and SQL dialect from the surrounding context (env vars, migration files, ORM models) before asking the user, drafts the query alongside a written rationale, and checks that the SQL parses in the right dialect, hits real columns, and stays read-only unless the user explicitly asked for a write.

OutcomeA reviewable SQL query plus a written rationale covering tables, joins, filters, output columns, and assumptions, all verified to parse in the inferred dialect, reference real columns from the schema, and avoid accidental writes.

Use when turning a meeting transcript into structured notes (a 5-bullet TL;DR, the decisions reached, and owner-attributed action items) that downstream automation can drop into a project tracker or recap email. Every owner, action, and decision is grounded in the actual transcript so the output stays trustworthy.

OutcomeA structured record of the meeting: a 5-bullet TL;DR, every decision with the quote that established it, and every action item with an owner and supporting quote pulled directly from the transcript.

Use when you have an idea for a data story and want a publication-quality chart back. The agent finds an authoritative public dataset, drafts three chart-type variants in parallel, picks the strongest, then iterates against a design checklist and a chart-design verifier until the chart actually carries the story.

OutcomeA publication-quality chart (PNG plus the Python that produced it), backed by an authoritative dataset, vetted against design fundamentals, and confirmed by a verifier to actually carry the data story.