@uplift-ds/workflow-skills
v1.2.0
Published
Uplift OS workflow package — align/prep/plan/build/verdict contracts, session/hat skills, and cross-agent sync tooling
Downloads
796
Readme
@uplift-ds/workflow-skills
Single source of truth for Uplift workflow contracts across Claude Code and Cursor.
What this package contains
- Quick-start / front door:
uplift/SKILL.md→/uplift(initialize a repo under the Uplift OS and surface the full capability set) - Session/hat operating context:
uplift-os/SKILL.md(the project-instructions equivalent for Claude Code — governs/session,/hat) - Canonical workflow commands:
commands/{align,prep,plan,build,verdict}.md+ session commandscommands/{session,hat}.md(first-class/sessionand/hat) - Workflow skills:
align/,prep/,plan/,build/,verdict/,conventions/,platform/(review/removed in 0.4.0 — useverdict/;domain/removed in 1.0.0 — per-repo defaults live inworkflow.config.json) - Agent definitions:
agents/done-checker.md(DoD judge) +agents/bug-hunter.md(correctness lens) +agents/security-reviewer.md(security lens, seated on security-sensitive diffs) — synced to.claude/agents/AND.cursor/agents/; the plan-cost reviewer panel/verdictspawns. No external review bots required — works in client-owned orgs; Bugbot is an optional accelerant on owned repos (orchestrator-triggered, once per review-ready PR) - Run conventions:
conventions/references/close-out-comment.md(every agent run ends with a structured Linear comment) +conventions/references/hydration-spec.md(the packet that wakes a fresh orchestrator brain) - Judgment session (the brain, UPLI-11):
orchestrator/scripts/judgment-session.mjs+orchestrator/judgment-prompt.md— wakes a fresh, toolless Fable via headless Claude Code with the five-part hydration packet; technical authority only; writes the unblock to Linear and the distilled rule to.ralph/memory.md - Stuck-helper:
verdict/scripts/stuck-check.mjs(deterministic escalation ladder: hint → reset → human) + opt-in hook templatetemplates/repo-adapter/claude-settings-stuck-hooks.json - Contract schemas:
contracts/*.schema.json - Manager-worker orchestration references:
orchestrator/ - Pattern decision matrix:
conventions/references/execution-pattern-classification.md - Repo adapter templates:
templates/repo-adapter/ - Sync CLI:
uplift-workflow-sync - Stage runner CLI:
uplift-workflow run <stage> <ISSUE_ID> --phase=preflight|postflight
Quick start in a new repo
pnpm add -D @uplift-ds/workflow-skills@latest
pnpm uplift-workflow-sync --force
# then, in Claude Code:
/uplift/uplift bootstraps the repo's Uplift Repo Scope block in CLAUDE.md (if missing), presents the Session Protocol kickoff menu, and shows the full capability surface — workflow commands, session/hat commands, the eight hats, and MCP status. This is the Claude Code analog of opening a Claude Chat / Cowork project that already has the OS project-instructions installed — except the operating context is delivered by this package and auto-updates on the next sync, instead of being hand-re-pasted.
Two layers, one package
As of 0.2.0 this package carries both layers of the Uplift Operating System and syncs both per-repo. They are distinct concerns but share one distribution + update mechanism:
| | Session/hat layer (uplift-os, uplift) | Workflow pipeline (align/prep/plan/build/verdict) |
|---|---|---|
| Governs | How work is bounded, hatted, and recorded | How one coding task is executed |
| Commands | /uplift, /session, /hat | /align → /prep → /plan → /build → /verdict |
| Source of truth | Notion (fetched at activation) | This package |
| State it touches | Notion Sessions DB | .ralph/<ISSUE_ID>/ artifacts, workflow.config.json |
Earlier designs kept the session adapter as a hand-placed user-level skill. From 0.2.0 it ships here instead, so a repo gets the operating context the same way it gets the workflow commands — pnpm add + sync — and it version-updates on the next bump (the Claude Chat adapter, by contrast, had no auto-sync and required manual re-paste).
Under the Engineering Architect hat, a session may drive the pipeline for a Linear issue (/align → /prep → /plan → /build) and record the outcome at /session close. The two command sets remain separate — neither is folded into the other — but they now travel together.
Cross-repo update model
- Update this package once.
- Publish a new version.
- In each consumer repo, bump package version and run sync.
No manual copy/paste of .claude/* or .cursor/* across repos.
Sync local wrappers in a consumer repo
Install package, then run:
pnpm add -D @uplift-ds/workflow-skills@latest
pnpm uplift-workflow-sync --forceBy default it writes:
.claude/commands/{align,prep,plan,build,verdict}.md.cursor/commands/{align,prep,plan,build,verdict}.md.claude/skills/*/SKILL.mdgenerated from package skills.cursor/skills/*/SKILL.mdgenerated from package skills.cursor/rules/workflow-pipeline.mdcworkflow.config.json(if missing, or overwrite with--force)
Run deterministic hooks
Use the runner around each LLM phase:
pnpm exec uplift-workflow run align SOC-415 --phase=preflight
# run reasoning using command/spec
pnpm exec uplift-workflow run align SOC-415 --phase=postflightOptional target directory:
pnpm uplift-workflow-sync --target=/path/to/repo --forceValidation
From repo root:
npm install
npm testPackaging strategy
- Canonical home:
UpliftDS/uplift-os(package at repo root). Extracted fromsocialite-mono/packages/workflow-skillson 2026-06-11 — the incubation this section used to describe is over. socialite-mono(and every other repo) is now a pure consumer:npm i -D @uplift-ds/workflow-skills+uplift-workflow-sync --force.- This repo runs under the OS it ships — its
.claude//.cursor/kit syncs from local source:node bin/sync.mjs --target=. --force.
