Shipd captures context unlike any other agent tool. You keep the specs, decisions and standing answers as plain, reviewable files in your repository and your workspace, not behind a login. Every part of shipd is a skill you can pick up on its own. The basics alone plan and build a change, and what they learn lands in a workspace any other skill can use later.
shipd is a collection of open-source skills built to the agentskills.io standard: together, a holistic end-to-end software delivery pipeline; apart, whichever ones you like. shipd is what you make of it.
Reads your codebase first, climbs the question ladder for whatever is left, and compiles intent into a plan, testable requirement deltas and a task list. Stops for your review; it never builds from a guess.
Runs the context gate, checks the plan has not been superseded, delegates every task to execution agents one tier down, then lets an independent validator refute each scenario before the semantic review gates the merge.
A nine-step walkthrough you drive one step at a time. Steps 1–7 explain the artifacts over a throwaway example, step 8 builds it for real, step 9 hands you your first command.
Decomposes an open question into bounded parts, searches each, reads the strongest sources and composes a report in which every claim carries a citation. An epic can link it.
Transcribes a screen recording on-device, grounds each spoken claim against the frame that was on screen, and resolves contradictions by recency.
Interrogates a product idea until it holds up, then installs a lint-clean workspace PRD through the engine's staged emit so epics can cite it.
An adversarial critic for an idea that is not planned yet. It reads the repo, pushes back instead of agreeing, names the strongest alternative and points at the skill that should pick the idea up. It changes nothing.
Breaks a feature too big for one PR into shared decisions, a design and a table of member changes with complexity ratings, then stops. Members are planned one at a time.
Authors or rewrites documentation against the shipd documentation standard, classifies each doc as concept, how-to or reference, and lints it clean. Docs only; nothing is shipped.
An ask-first agent answers one compact decision from your standing positions, with citations, before any person is interrupted. What it cannot answer is queued for a human; the run never blocks.
Distills this repo's spec artifacts and answered queue entries into the workspace wiki, interviewing you only on the gaps and contradictions the scan surfaced.
Captures a durable preference into the personal memory store through one confirmed write. Every future plan consults it before asking you.
List every preference the store holds, or remove one after a single confirmation: the read and delete counterparts to remembering.
A context-sufficiency gate checks the plan against the real codebase before any code is written. Insufficient context parks the plan as rejected for enrichment; the system never builds on a guess.
Installs the managed files, requires the semantic-review check on the repository and enables auto-merge, taking your consent before anything changes on GitHub.
Reports a change's lifecycle position, validates its structure, or runs a guarded transition that asks before forcing past a guard.
A read-only preflight reporting ok, warn or fail per check. Proposes one remedy per remediable finding and runs only what you consent to, then re-runs and reports before and after.
Structurally validates specs and change deltas: requirement grammar, scenario coverage and delta consistency against the capability library.
An independent validator spawned by build tries to refute every scenario in the spec against the real, running code, and returns one verdict per scenario. Refuted goes back; only confirmed moves on.
A semantic review of local changes against a base ref: files grouped into cohorts, changed signatures chased to their call sites, findings rated high, medium or low, and a ship-it or fix-required verdict posted to the PR.
Retrieves the related specs, reproduces the failure and repairs the code that drifted from documented behaviour, with a regression test. Never edits a spec, never opens a PR.
Drives a real browser to operate your app and verify a change against it: probes the live DOM before authoring a selector, waits for a named completion signal, and ends on a pass or fail verdict backed by console and network evidence.
Records a branded demo video of a change, through the browser for a web app or a terminal tape for a CLI, and stops on a missing tool rather than attempting a doomed recording.
Creates, inspects, clones or syncs a portable workspace: a manifest plus job wiki whose member repos materialize on this machine via the sync ladder of worktree, reference-clone or full clone.
Authors and registers the setup scripts every fresh worktree runs, so a new change branch starts with its dependencies and environment in place.
Authors a lint-clean outcome brief from a workspace-first interview, reports every initiative's requirement progress, reviews a brief's outcomes, or attaches one to an epic.
One human approval, the epic, and every unplanned member is driven plan, gate, build and merged PR in risk-ascending order, one worktree and branch each. A parked member pauses only itself.
Reads an epic through the engine and explains what it is for, the decisions holding it up, how its members compose and where delivery stands, with a diagram only where one carries structure.
The full-screen delivery board: what is in flight, what is parked, what merged this week and which agent is doing what right now.
In-flight changes across the repo root and every worktree, plus throughput metrics derived from the immutable archive.