Ship solo, safely.
Soloship is guardrails for non-coders building software through AI agents. It supports Claude Code, Codex, and Google Antigravity as guardrail targets (plus environment-sync skills that set up Cursor and Grok Build): mechanical enforcement that fires automatically (29 hook protections, 19 always-on rules, and CI checks, no judgment calls required), 51 workflow skills drawn from Soloship's own work plus five best-in-class upstream projects (Compound Engineering, Superpowers, Impeccable, gstack, ui-ux-pro-max, full attribution below), each with enforcement gates and anti-rationalization tables so the agent can't cut corners, and a one-command setup that detects your stack and wires guardrails into the project.
Everything ships inside the one Soloship plugin. Nothing depends on other plugins being installed.
Quick reference: aifoundationlevels.com/soloship-cheatsheet
You need the agent you want to use and Node.js 20 or newer:
--agent antigravity).npx soloship init, upgrade, doctor, and rollback.You don't need to install anything from npm by hand. npx downloads soloship on demand and caches it. Soloship is live on npm, verify with npm view soloship.
| Surface | What it gives you | Install/update owner |
|---|---|---|
| Codex plugin | Soloship skills for audit, bootstrap, plan, implement, review, browse, and ship workflows | codex plugin ... |
| Claude Code plugin | /soloship:* slash commands backed by the same skills/ source tree | /plugin ... in Claude Code |
| npm CLI | Project guardrails: docs, rules, Claude hooks, CI, rollback stamp | npx soloship ... |
| Antigravity guardrails | .agents/rules/ + .agents/hooks.json + the global Gemini plugin (npm run antigravity:install-local) | npx soloship init|upgrade --agent antigravity |
Use the plugin for daily workflow. Use the npm CLI once per project, then again when guardrails need refreshing.
Use one of these paths.
Codex only, if you work in the Codex app or Codex CLI:
codex plugin marketplace add thedigitalorganizer/soloship
codex plugin add soloship@soloship
npx soloship init --agent codex
Start a new Codex thread after the plugin install. The project setup creates AGENTS.md and .codex/rules/; it does not create Claude hooks.
Claude Code only, if you only work in Claude Code:
/plugin marketplace add thedigitalorganizer/soloship
/plugin install soloship@soloship
Then set up each project:
npx soloship init --agent claude
This creates CLAUDE.md, .claude/rules/, and .claude/settings.local.json.
Both Codex and Claude Code, if you switch between them:
codex plugin marketplace add thedigitalorganizer/soloship
codex plugin add soloship@soloship
/plugin marketplace add thedigitalorganizer/soloship
/plugin install soloship@soloship
Then set up each project once:
npx soloship init --agent both
Both agents use the same skills/ source and the same docs/plans/ and docs/solutions/ project artifacts. Claude Code gets .claude/ hooks/rules; Codex gets .codex/rules/ and AGENTS.md.
Add the Soloship marketplace, then install the plugin:
codex plugin marketplace add thedigitalorganizer/soloship
codex plugin add soloship@soloship
Start a new Codex thread after installing or updating so Codex loads the refreshed skills.
Inside Claude Code, run these one at a time:
/plugin marketplace add thedigitalorganizer/soloship
/plugin install soloship@soloship
Once installed, every /soloship:* slash command is available in Claude Code.
A laptop /plugin install is user-scoped and never carries over to Claude
Code on the web — that's documented behavior,
not a bug. The fix is repo-scoped enablement: as of v0.24.0, npx soloship init and npx soloship upgrade write extraKnownMarketplaces +
enabledPlugins into the project's checked-in .claude/settings.json. Commit
that file and every cloud session on the repo auto-installs the Soloship
plugin at session start (network access to github.com required in the
environment). Keep your laptop install too — the scopes coexist cleanly, and
anyone who wants out on one machine sets "soloship@soloship": false in their
gitignored .claude/settings.local.json.
For projects you don't want to re-run the CLI in, the snippet is:
{
"extraKnownMarketplaces": {
"soloship": { "source": { "source": "github", "repo": "thedigitalorganizer/soloship" } }
},
"enabledPlugins": { "soloship@soloship": true }
}
If you previously installed Superpowers, gstack, or Compound Engineering as separate plugins: Soloship vendors the curated set from each (see Skill architecture), so they can be removed once Soloship is installed — fewer plugins to update, one namespace, no dispatch collisions. On each machine:
/plugin uninstall superpowers
/plugin uninstall gstack
/plugin uninstall compound-engineering
then remove their marketplaces from /plugin marketplace list if no other
plugin uses them. Anything you miss from the full upstream sets can be
reinstalled side-by-side later — Soloship's commands are namespaced and won't
collide.
Open the project you want to use Soloship in and run one setup path.
For Codex:
npx soloship init --agent codex
For Claude Code:
npx soloship init --agent claude
For both:
npx soloship init --agent both
init creates the documentation structure, 19 workflow rules, CI checks, the automation registry scaffold, the .soloship/version stamp, and the right agent-facing guidance. Claude installs hooks under .claude/settings.local.json; Codex installs rules under .codex/rules/ and uses AGENTS.md; Antigravity installs .agents/rules/ and .agents/hooks.json. Codex hooks are intentionally not ported yet; they wait on verified Codex hook payloads.
For an existing codebase, run Soloship audit before bootstrap/setup so the guardrails match the code instead of guessing. In Claude Code, use /soloship:audit; in Codex, invoke the Soloship audit skill.
Update each surface separately:
# Project guardrails
npx soloship upgrade --agent codex
npx soloship upgrade --agent claude
npx soloship upgrade --agent both
# Codex plugin
codex plugin marketplace upgrade soloship
codex plugin add soloship@soloship
# Claude Code plugin
/plugin update soloship@soloship
docs/SOLUTION_GUIDE.md is generated from the solution-doc schema and carries a
version marker. When that schema changes, init and upgrade report your copy
as stale rather than overwriting it — your project may have extended it. To
take the new version (the previous copy is kept as .bak):
npx soloship init --refresh-guides
npx soloship upgrade --refresh-guides
For local Codex dogfooding while developing this repo:
npm run codex:sync-local
codex plugin add soloship@personal
Start a new Codex thread after any plugin reinstall.
Both agents use the same repo, the same skills/ source tree, and the same plan/solution artifacts under docs/. Claude Code reads CLAUDE.md, .claude/rules/, and .claude/settings.local.json. Codex reads AGENTS.md and .codex/rules/. Claude uses /soloship:*; Codex uses the installed Soloship plugin skills.
Soloship didn't start as a project. It started as a frustration.
I had 40+ solution documents in docs/solutions/: real fixes to real problems I'd hit before. The kind of knowledge that should prevent making the same mistake twice. But the agents kept making the same mistakes. I finally stopped and asked why.
The answer: none of the tools in my stack were reading the solutions at the point of execution. Compound Engineering's planning workflow dispatches a learnings-researcher agent that searches docs/solutions/, but it runs once, during the research phase, and its results get compressed into a few bullet points in the plan doc. The execution workflow never touches the solutions directory. It reads the plan, follows the plan, and trusts the plan to carry everything forward. With 40+ solution docs, most get scored out during the research pass. What survives is a summary of a summary. And once execution starts, there's no retrieval, even when the agent is hitting the exact error a solution doc describes.
So I tried rules: a solution-search rule that told the agent to check solutions before planning, debugging, or reviewing. It helped, when the agent followed it. But rules are suggestions. The agent would rationalize past them: "this is a simple change, no need to search." I'd ask if the rules were followed, and the honest answer was usually no.
So I tried hooks: mechanical triggers that fire regardless of what the agent is paying attention to. Hooks can't be rationalized away. That worked better. But wiring up hooks, rules, solution docs, AGENTS.md files, and CI checks for each new project was its own overhead.
And the workflow skills I was using weren't quite right either. gstack has a strong skill set (QA, design review, security, shipping), but its skills are verbose and try to cover too much surface area. The agent would get lost in instruction volume, or the skill would prescribe steps that didn't apply to a solo operator's workflow. So I started pulling the good parts out and writing leaner versions: opinionated for the solo use case, with enforcement gates and solution-search wired in from the start. The skills worth keeping as-is got vendored directly into Soloship with full attribution, so one plugin install carries everything and nothing breaks when an upstream plugin isn't installed.
That became Soloship.
Three layers, from most mechanical to most guided:
Hooks fire automatically in Claude Code sessions. They can't be rationalized away. Dangerous command blocking, security scanning on commits, auto-lint, CHANGELOG enforcement, plan validation, deploy discipline, the billing and recurrence gates, session presence for parallel agents, checkpoint commits, and a reply timestamp on every response so session logs can reconstruct when work actually happened. If the agent forgets, the hook remembers. Codex hook parity is intentionally deferred until Codex hook payloads are verified.
Rules are injected into every agent session as always-on context. Solution search before planning, plan materialization after plan mode, plan rationale and claim-verification requirements, plan lifecycle enforcement, a QA plan in every plan, the billing-confirmation and recurrence gates, parameterize-constants, deploy-from-main-only, the automation registry, and a browser-QA gate that blocks "done" until the change has been exercised in a real browser (with a test account where a flow needs login). The agent can't not see them, even if a skill doesn't reference them.
Skills are guided workflows. Claude Code exposes them as /soloship:* commands; Codex exposes them through the installed Soloship plugin. Each adds enforcement gates (checklists the agent must complete) and anti-rationalization tables (preemptive counters to the ways agents cut corners).
Soloship runs the same skill set under two postures, decided by the session's model (the auto-loaded model-mode rule):
docs/reports/2026-08-11-fable-brief-ab-experiment.md).Heavy skills carry an explicit "Model posture" section naming their own gates; every other skill inherits the rule's generic definitions. Hooks are mechanical and identical in both postures. /soloship:fable is the entry point for launching Fable-shaped work: it qualifies the task (routing routine work back to Opus — the budget gate), builds the goal brief, launches, and harvests lessons.
Every one of the 47 skills ships inside the Soloship plugin. Earlier versions routed to external plugins (Superpowers, Compound Engineering, gstack) when they were installed; that design was retired in v0.5.0. The curated set is now vendored directly into Soloship, so there are no external plugin dependencies and no broken dispatches for users who only install Soloship.
Skills come in three kinds:
Soloship-native (19): the logic was written for Soloship. audit, bootstrap, brainstorm, cleanup, cron, debug, design-review, fable, finish, grill-me, implement, learn, onboard, plan, review, shipfast, shipthorough, spec, status. Several of these embed methodology adapted from the upstream projects (for example, plan runs a Compound-Engineering-derived plan-writing flow and debug runs the Superpowers 4-phase discipline), with Soloship's enforcement gates layered on top.
Vendored review and design (15): the gstack plan-review set (ceo-review, eng-review, devex-review, plan-design-review, autoplan), the Compound Engineering review set (code-review, deepen-plan, document-review), the Impeccable design set (clarify, critique, polish, simplify, frontend-design, ui-audit), and ui-ux-pro-max.
Vendored discipline and utilities (12): the Superpowers discipline skills (executing-plans, subagent-driven-development, test-driven-development, using-git-worktrees, verification-before-completion, writing-plans) and the gstack standalones (browse, qa, cso, context-save, context-restore, office-hours).
All 45 are invoked the same way: /soloship:<name>. Source attribution lives in each skill's header and in THIRD_PARTY_NOTICES.md.
When you run /soloship:bootstrap (or /soloship:audit then /soloship:bootstrap on an existing project), it detects your language, framework, and package manager, then installs:
docs/plans/, docs/solutions/, docs/audit/, docs/automations/ (the automation registry), AGENTS.md stubsstatus: done while the plan body still has unchecked boxes or PENDING/BLOCKED/IN PROGRESS markers). PostToolUse (7): auto-lint, CHANGELOG check, recurrence audit, live-data evidence check, duplicate-component warning, session heartbeat, browser claim. Stop (4): plan validation + workflow navigator + handoff reminder (the plan-status contradiction check reads the plan body, not just frontmatter — it prompts a review instead of commanding status: done when open items are still listed), a reply timestamp (so session logs can reconstruct when work actually happened), a browser-teardown reminder, and a deploy-lock reminder. SessionStart (3): checkpoint commit, update check, session presence. SessionEnd (2): browser-claim release, deploy-lock release.CLAUDE.md, AGENTS.md, CHANGELOG, SOLUTION_GUIDE, sized to your stackRun bootstrap once per project. For existing code, run /soloship:audit first so bootstrap can tailor the setup.
19 Soloship workflow skills invoked as /soloship:* slash commands. Each one handles orchestration, enforcement, and artifact contracts. One of the 19 (/finish) is a faithful vendor of a Superpowers skill, renamed to fit the Soloship slash-command surface.
Setup & orientation
/soloship:audit: Deep 2-phase codebase investigation. Phase 1 launches 4 parallel agents to map architecture, conventions, decisions, and infrastructure. Phase 2 launches 7 more to assess quality, entanglement, security, dependencies, gaps, leverage points, and the automation surface (every cron, webhook, and scheduled job plus its monitoring state). Human checkpoint between phases prevents building assessment on wrong assumptions. Produces docs/audit/AUDIT-YYYY-MM-DD.md + audit-findings.json./soloship:bootstrap: Configures governance from audit findings or interactive questions. Creates CLAUDE.md and/or AGENTS.md guidance, installs the 19 workflow rules, seeds the automation registry, and wires up Claude hooks when Claude is targeted. Never overwrites existing files. Anti-rationalization table blocks "I'll set up governance later."/soloship:onboard: Reads CLAUDE.md, AGENTS.md, audit reports, and recent git history to produce a 7-section orientation briefing. Flags stale audit reports. Fully self-contained.Daily work
/soloship:brainstorm: Feature exploration that merges Compound Engineering's brainstorm methodology with Superpowers' brainstorming discipline. Ends with a mandatory design-first nudge: sketch before you plan. For demand-validation questions (should this exist?), use /soloship:office-hours instead./soloship:grill-me: Relentless pre-plan interview that walks every branch of the design tree until user and agent share a complete mental model. Refuses to produce a plan or any code until alignment is explicit. Triggered explicitly ("grill me", "interview me") or by /soloship:plan on medium-to-large work. Adapted from Matt Pocock's grill-me (MIT)./soloship:spec: Writes formal specifications with numbered acceptance criteria, data models, API contracts, user flows (including error states), and explicit out-of-scope boundaries. 8-point verification checklist. Fully self-contained./soloship:plan: Searches docs/solutions/ for prior art, reads architecture context, then runs the Compound-Engineering-derived plan-writing methodology. Every plan must include a QA Plan section with the verification method matched to each touched surface. The enforcement gate validates: Why lines, Key Decisions, Execution Strategy, Handoff section, QA Plan, no unaddressed pitfalls, and that non-trivial work was preceded by /soloship:grill-me. Review is separate, handled by /soloship:review./soloship:implement: Finds the most recent plan in docs/plans/, claims it (so parallel sessions don't double-work it), sets up an isolated worktree, then runs the Compound-Engineering-derived execution methodology with branching and quality checks. Freshness check warns on stale plans. A mandatory QA Gate (Step 2.6) executes every row of the plan's QA Plan before completion, and browser QA via /soloship:browse is required for any user-facing flow, with a test account where a flow needs login. Defaults to a local merge into the base branch when execution finishes; it does not auto-create a GitHub PR. Ask explicitly ("open a PR for this") or use /soloship:finish Option 2 if you want one./soloship:debug: Iron law: no fixes without root cause investigation. Searches solutions for prior art first, then runs the Superpowers-derived 4-phase discipline (Investigate, Analyze, Hypothesize, Implement). Nudges /soloship:learn for non-obvious fixes./soloship:learn: Captures knowledge from non-obvious work. Runs the Compound-Engineering-derived compound methodology to write a solution doc. Adds Soloship protocols: JSONL logging for cross-session search, architecture registry drift checking, and distributed AGENTS.md propagation (pitfalls into existing AGENTS.md files, new ones for directories above the 3-file governance threshold). Anti-rationalization table blocks "this fix was straightforward, not worth documenting."/soloship:cleanup: Knowledge system maintenance. Launches 5 parallel audit agents (solution health, overlap detection, plan lifecycle, AGENTS.md staleness, index sync), presents findings interactively, then executes approved changes in a single atomic commit. Merge candidates require 2-of-3 signal threshold./soloship:component-inventory: Scans the codebase for UI components (React/Vue/Svelte) and generates docs/architecture/COMPONENTS.md — name, file, purpose, props, and usage sites per component, plus a "Possible duplicates" section the user decides on. Delta-updates an existing inventory (unchanged rows are preserved byte-for-byte, so a no-change re-run produces no diff). The data source behind the component-reuse rule and the duplicate-component warn hook.Operations
/soloship:status: One read-only dashboard for "what's going on in this project": active agent sessions (who's working where), the plan board (every plan by status with progress), and deploy state (what's live vs what's merged and waiting)./soloship:cron: The management console for every automation a project owns (cron jobs, scheduled workers, local launchd/crontab jobs, webhooks). Reads the automation registry, queries the watchdog, live-discovers unregistered jobs, troubleshoots anything dark, and enforces the build-time contract for new automations: register, deploy, wire the check-in, observe the first check-in.Shipping
/soloship:finish: When implementation is done, all tests pass, and you need to decide how to integrate the work. Walks the merge / PR / cleanup options with a structured decision tree. Faithful vendor of Superpowers' finishing-a-development-branch skill, renamed for the Soloship slash surface./soloship:shipfast: Emergency deploy pipeline. Lint (with auto-fix tolerance), test (pre-existing failures allowed), build (must pass), commit, push, deploy. Auto-detects platform. Even shipfast shows the deploy manifest and asks once. Minimum viable safety, maximum speed./soloship:shipthorough: Full due diligence: preflight checks, base branch merge, lint, test, inline quality gate (TypeScript + linter + dead code + shellcheck), coverage audit, 3-pass code review (via /review), registry update, CHANGELOG enforcement, plan lifecycle cleanup, bisectable commits, local merge into the base branch and push (PR only on explicit opt-in), verification gate, deploy.Quality
/soloship:review: Detects whether the target is a plan or code. Plans go to eng-review, ceo-review, plan-design-review individually, or autoplan for all four (adds DX review) in one auto-decided pass. Code goes to code-review for PR-scale multi-agent analysis, or three inline passes (structural, adversarial, design slop lens) for quick local checks./soloship:design-review: Two-pass visual audit. Pass 1 covers spacing, hierarchy, and consistency (gstack-derived). Pass 2 is Soloship's own AI slop detection (inspired by Impeccable): flags generic gradients, default shadows, "Welcome to" copy, 3-column feature grids, and other patterns that mark AI-generated design. Each fix committed atomically with before/after screenshots.Plus 27 vendored skills invoked the same way: /soloship:browse (headless browser for QA), /soloship:qa, /soloship:autoplan, /soloship:cso, /soloship:context-save / /soloship:context-restore, /soloship:office-hours, the four plan-review skills, the CE review set, the Impeccable design set, the Superpowers discipline set, and /soloship:ui-ux-pro-max. See Skill architecture for the full breakdown.
See the Install section above for the Codex and Claude install commands. Once installed, the daily flow inside any project is:
New project:
/soloship:bootstrap # set up the guardrails
/soloship:brainstorm → /soloship:plan → /soloship:implement
Existing project:
/soloship:audit # understand what's there first
/soloship:bootstrap # set up guardrails tailored to what audit found
/soloship:plan → /soloship:implement → /soloship:shipthorough
/soloship:bootstrap calls npx soloship init under the hood, so you don't usually need to think about the npm CLI. But it's there if you want to script setup, run it in CI, or skip the slash command entirely. Run any of these from your project root:
npx soloship init --agent both # initial setup: creates docs, rules, CI, AGENTS.md, and Claude hooks
npx soloship upgrade # refresh hooks, rules, and the .soloship/version stamp
npx soloship doctor # check Claude Code, Codex, and project guardrail status
npx soloship rollback # restore the last safety snapshot
npx auto-downloads the latest Soloship from npm the first time you run it, and caches it after that. There's no separate npm install -g step.
/soloship:brainstorm nudges you toward visual design before /soloship:plan. Design catches problems text can't.| Phase | Status | What it delivered |
|---|---|---|
| 1-2 | Done | Cleanup + npx soloship init with stack detection |
| 3-4 | Done | /audit + /bootstrap skills |
| 5-6 | Done | The initial skill set + hooks + rules |
| 7 | Partially done | The safety floor shipped (Semgrep security scanning, checkpoint/rollback, phone-a-friend warnings). Surface simplification has not started. |
| 8 | Not started | Graduation system, methodology documentation |
Since the phase plan was written, releases 0.5 through 0.13 also vendored the full 45-skill set into the plugin, added Codex support, the browser QA gate, the QA-plan requirement, deploy discipline (deploy-from-main-only, the deploy lock, the prod tag), cross-session coordination for parallel agents, and the automation registry with its one-watchdog standard. See CHANGELOG.md for the full history. Phase 8's graduation system (calibrated thresholds that tell you when your project has outgrown solo mode) remains open.
Soloship curates and vendors skills from five outstanding Claude Code plugin ecosystems. One Soloship plugin install per host gives users the curated set; full credit and install links for the authors stay here. Full attribution and version pins live in THIRD_PARTY_NOTICES.md.
Compound Engineering, Kieran Klaassen (Every). The brainstorm, plan, work, compound loop is the spine of how Soloship thinks about engineering. /review inherits CE's multi-agent review pattern.
Superpowers, Jesse Vincent. The discipline skills: systematic-debugging's "no fixes without root cause," verification-before-completion's "evidence before claims," test-driven-development, writing-plans, executing-plans, subagent-driven-development, using-git-worktrees, finishing-a-development-branch (renamed to finish in Soloship, invoke via /soloship:finish), and brainstorming. Nine skills total.
Impeccable, Paul Bakaus (extending Anthropic's original frontend-design). Design vocabulary and steering commands that let non-coders ship work that doesn't look AI-generated. Soloship vendors frontend-design and five /i-* commands; 12 more are in the full plugin.
ui-ux-pro-max, nextlevelbuilder. The design reference library: styles, palettes, font pairings, UX guidelines, chart patterns, across every stack the agent might target.
gstack, Garry Tan (YC). The solo builder toolkit. Soloship vendors 12 of the most non-coder-friendly skills: autoplan (chains CEO + design + eng + DX reviews), context-save / context-restore (the checkpoint pair), browse (headless browser daemon, re-vendored with Soloship-native paths so it works standalone), qa, design-review, the four plan-review skills (eng, ceo, design, devex), office-hours, and cso. Full gstack has ~25 more.
If any of these are useful to you, please install the full upstream plugin. Each THIRD_PARTY_NOTICES.md section has the one-line install command, and you'll get everything the author built, not just Soloship's selection.
Also influential but not vendored:
intent-layer (crafter-station/skills, built on The Intent Layer by Tyler Brandt). /learn Steps 4-5 adapt the concept of distributed per-directory AGENTS.md files for codebase navigation. Soloship's version is continuous (updates on every /learn pass, not one-shot), threshold-gated (3+ source files before creating), append-only with dated attribution, and scoped to solution-doc evidence rather than speculative. /bootstrap and /cleanup also maintain the AGENTS.md network.
Serena, symbol-level LSP code navigation. Optional. Worth adding once a codebase outgrows file-level tools; see Serena's README for install instructions.
The broader design traces back to a research pass across: Ousterhout on strategic vs tactical programming (you are the architect, the agent implements), Hickey on simple vs easy, Metz on dependency awareness and sizing rules, Meadows on leverage points in systems, the BCG "AI Brain Fry" finding that productivity drops past three tools, Kathy Sierra on the collapse zone (only automated process survives when things break), and the Codified Context paper that validated the CLAUDE.md + AGENTS.md + docs/ three-tier pattern.
.claude-plugin/ # Claude Code plugin manifest (plugin.json, marketplace.json)
.codex-plugin/ # Codex plugin manifest
.agents/plugins/ # Codex marketplace entry
bin/soloship.js # CLI entry point
src/ # TypeScript source for the installer
cli.ts # Commander CLI definition
init.ts # Main init orchestration
detect.ts # Stack detection
scaffold.ts # Folder + doc creation
hooks.ts # Claude Code hook configuration
rules.ts # Claude and Codex workflow rule installation
ci.ts # GitHub Actions + architecture fitness
templates.ts # CLAUDE.md / AGENTS.md / CHANGELOG / SOLUTION_GUIDE generators
skills/ # Shared Claude Code and Codex skills shipped by the plugins (46 total)
# All skills are invoked as /soloship:<name>. Source attribution lives
# in THIRD_PARTY_NOTICES.md; no source prefixes leak into command names.
#
# There is deliberately NO commands/ directory. Claude Code resolves commands
# and skills in ONE namespace, so a command file sharing a skill's name
# shadows it and the workflow becomes unreachable. Removed in v0.21.0;
# validate-plugin-metadata.js blocks reintroduction.
# Soloship-native workflow skills (20):
audit/ bootstrap/ brainstorm/ cleanup/ component-inventory/ cron/ debug/
design-review/ fable/ finish/ grill-me/ implement/ learn/ onboard/ plan/
review/ shipfast/ shipthorough/ spec/ status/
# Plan-review skills, derived from gstack (5):
ceo-review/ eng-review/ devex-review/ plan-design-review/ autoplan/
# Code-review, design, and frontend skills, derived from CE and Impeccable (8):
code-review/ deepen-plan/ document-review/ clarify/ critique/
polish/ simplify/ frontend-design/
# Discipline skills, derived from Superpowers (6):
executing-plans/ subagent-driven-development/ test-driven-development/
using-git-worktrees/ verification-before-completion/ writing-plans/
# Standalone vendored utilities (8):
browse/ # gstack's headless browser daemon, re-vendored with
# Soloship-native paths so it works without gstack
# installed. Compiles its launcher on first use
# (build-on-host for arm64/x86_64 portability).
cso/ qa/ context-save/ context-restore/ office-hours/ # from gstack
ui-audit/ # from Impeccable
ui-ux-pro-max/ # from nextlevelbuilder
references/ # Shared checklists (a11y, code review, perf, security, testing)
vendored/ # Per-source LICENSE / NOTICE / VERSION / README (attribution archive)
MIT.
Soloship is opinionated and unfinished. It's the working toolkit of one person who builds software through AI agents and is trying to do it responsibly. If you find it useful, great. If you think a decision is wrong, open an issue. The adversarial-review phase of this project already taught me that the things I'm most confident about are the things most likely to need a second pair of eyes.
FAQs
Systematic programming methodology for AI-assisted development
The npm package soloship receives a total of 22 weekly downloads. As such, soloship popularity was classified as not popular.
We found that soloship demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Product
Socket Threat Intel now shows which malware campaigns were blocked or detected in your organization, with links to scans and Firewall events.
Security News
Fixes for eMudhra and Brotli are out. Getting them everywhere they’re needed means tracking down bundled libraries, runtimes, and certificate stores.
Threat Research
/Security News
Socket uncovered secret-stealing workflows that ran in LiteLLM’s GitHub repository 11 days before the supply chain attack on PyPI and Docker.