/suede-ship · multi-agent shipping DAG

Fifty agents. One graph. One shipped change.

Suede Ship runs a nontrivial change as a graph instead of a chain: scout, multi-lens research, a gap critic that names what went unread, a lane plan with explicit file ownership, disjoint parallel build, adversarial refutation of every finding, and an integration gate on typecheck, build, and tests. It halts on a blocking hazard or a lane collision rather than plowing through, and it reads production without ever deploying.

View skill folder Install options

"ship this change across the whole surface" /suede-ship Ask in plain words. The router reads the request and loads this lane; MCP agents find it with list_suede_skills.

$ /suede-ship add rate limiting to the api scout: clean · 4 lanes planned · no collisions build 4/4 · refute killed 2 of 9 findings GATE: TYPECHECK + BUILD + TESTS PASS handoff written · not deployed

Public install commands

Claude Code full pack

/plugin marketplace add JasonColapietro/suede-creator-skills
/plugin install suede-skills@suede

Claude Code project-level copy

git clone https://github.com/JasonColapietro/suede-creator-skills.git /tmp/suede-creator-skills
mkdir -p .claude/skills
cp -R /tmp/suede-creator-skills/skills/suede-ship .claude/skills/

Codex installer

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo JasonColapietro/suede-creator-skills \
  --path skills/suede-ship

What it covers

Scout the repo first: origin state, dirty files, sibling worktrees, and deploy-time landmines, as a manifest only.
Sweep with multi-modal research lenses that each search a different way, where every claim carries a file:line, sha, PR, or doc url.
Run a completeness critic that names what went unread, then one bounded round to fill the gaps.
Plan lanes with explicit file ownership, so no two parallel workers can claim the same file.
Refute every finding adversarially on fresh context, where a majority of skeptics kills it.
Gate on a real barrier: typecheck, build, and tests against the integrated worktree, then write an evidence handoff.

When to use it

Any nontrivial change that touches more than one file or surface and deserves surgical, research-heavy fan-out.
Work in a repo where several sessions run at once and a lane collision would otherwise clobber a sibling branch.
Changes where you want the findings refuted before you read them, not after you ship them.
Runs you want resumable, replaying unchanged agents from cache instead of paying for the whole graph again.
Not for: high-volume, well-specified work that splits into independent worker-sized tasks (use suede-codex-fleet, which bills to the OpenAI subscription instead).
Not for: findings-only review with no code change (use suede-code-review).
Not for: wiring CI and branch protection (use suede-ci-gate).

Best prompt

Use $suede-ship on ~/code/my-app to add per-account rate limiting to every API route. Keep my scope verbatim so the planner can split it into lanes, and state the cost before launching.

Related skills

Related skills from this docs set: Suede Fable Fleet, Suede Agent Teams, Suede Code Review.