workflows

# Workflows Skills compose into repeatable **chains** — one skill does one thing; a workflow runs several in order. The router for any developer or researcher: match the job to a chain, state it, run it. `Chains.md` carries every exact skill sequence; in this repo the chains are also an **OKF bundle** (`workflows/index.md`). **Invoked without a named workflow** → present both tables below and ask which to run; **with one named** → state its chain and run it. ## All workflows | Family | Job | Workflow | |--------|-----|----------| | ship | Ship fast, scope clear enough, explore mid-flight | ship-fast | | ship | Spec-driven build, or a bug fix | spec-to-ship | | ship | Learn a domain / library / API before building | research-to-build | | ship | Deliver cited findings + visuals, not code | research-to-report | | ship | Compare options before buying or adopting | research-to-buy | | ship | Get productive in an unfamiliar repo | onboard | | ship | Build a TypeScript CLI | build-cli | | ship | Build an MCP server | build-mcp | | ship | Quick UI polish (anti-slop) | ui-feature | | ship | Full frontend feature end-to-end | frontend-build | | ship | Backend / API change, security-gated | api-feature | | ship | Harden existing code before release | security-pass | | ship | Cut context-window bloat | context-diet | | ship | Clean up config / project repo | repo-hygiene | | library | New skill, clear single capability | new-skill-quick | | library | New skill, complex / shape unclear | new-skill-heavy | | library | Fix a skill's broken structure | canonicalize-skill | | library | Skill won't trigger or mis-fires | fix-trigger | | library | Improve a target hands-off (loop) | autonomous-loop | | library | Whole-library health audit | library-audit | | library | Re-tune skills for a new model | model-upgrade | | library | Build / refactor many skills at once | batch-build | | library | Cut a tagged release | release | **Running a chain:** state the chain to the user, then **invoke each step — they are skills, not prose**: load each named skill with the Skill tool (or `Skill:Workflow` for one step of a skill) and let it do the work, in order. Honor gates (e.g. `Verify` before shipping) and any loop's named exit condition. ## Gotchas - Every chain step names a **skill**, never a host slash command — that is what keeps a chain runnable on pi, opencode, and Claude Code alike. Old command equivalents: `/plan`→`Spec`, `/build`→`Build`, `/test`→`Test`, `/capture`→`Reflect`, `/verify`→`Verify`, `/commit-push-pr`→`GitHubOps:CommitPush`+`GitHubOps:PullRequest`. - Every autonomous loop names its **exit condition** first; no unbounded loops. `Iterate` already calls Verify+Reflect and `Orchestrate` gates with Verify — don't re-chain those by hand. ## Examples - "ship a small feature fast" → ship-fast → state chain → `Spec` → `Build` → … → `GitHubOps:CommitPush` → `GitHubOps:PullRequest`. - "my skill won't trigger" → fix-trigger → OptimizeDescription → TestSkill.