Savvy Flow

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, nothing to install first. The skill is a Claude Code orchestration flow: it is invoked manually (/savvy-flow <task>, model invocation is disabled in its frontmatter) and it only works when its five worker agent definitions are installed with it, which this install does. Using it afterwards spends real model usage: the workers run on Fable and Opus tiers inside my own session and subscription. Install "savvy-flow" for me. It is an orchestrated delivery flow: the session model plans, owns design, delegates implementation to five tiered worker subagents (savvy-fable, savvy-heavy, savvy-careful, savvy-medium, savvy-light), and adversarially reviews what they return before accepting anything. Repository: https://github.com/JohnnyVizz/claude-kit (the skill is the plugins/savvy-flow/skills/savvy-flow/ folder, and its required worker agents are plugins/savvy-flow/agents/, both at the pinned commit) Pinned version: install the code at commit c871c153ead54ad9547cc717d0f3e5a676c73986 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/JohnnyVizz/claude-kit into a temporary folder, check out commit c871c153ead54ad9547cc717d0f3e5a676c73986, and copy the plugins/savvy-flow/skills/savvy-flow/ folder and the plugins/savvy-flow/agents/ folder from that checkout. Then compare every file with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - skills/savvy-flow/SKILL.md: SHA-256 71356eefe21aca9f4afc3ed28c0a808495160ebe52fce4003c527e7b72863cdb - agents/savvy-careful.md: SHA-256 ddd5169be679724addde137c1a620bcf1d14b9bf418ec3af1557829fea4ef4e2 - agents/savvy-fable.md: SHA-256 52429b8980a104ce2ad95c46f040187f83946ab52db750eed6143c8d3cf82985 - agents/savvy-heavy.md: SHA-256 9250b53b63970f55d07ef4972c30ff10c27abff78264a76610dbe530e02f6414 - agents/savvy-light.md: SHA-256 47f1f4c25de599592490f4ca69f5e8f97b4d6fa3d0a1a74a53427288620e0f35 - agents/savvy-medium.md: SHA-256 646d3fe6be47f062b4d263794548d094bc5c53caedcce79b5894f6a254a8f81c If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full and the five agent files, then tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The install has two parts, exactly as the upstream plugin ships them (plugin "savvy-flow", version 1.2.0): the skill itself, one file (skills/savvy-flow/SKILL.md, the orchestrator method), and the five worker agent definitions in agents/ (savvy-fable on Fable at high effort for the hardest problems, savvy-heavy on Opus at xhigh effort to understand and find, savvy-careful on Opus at high effort to do delicate work right, savvy-medium on Opus at medium effort for standard work, savvy-light on Opus at low effort for mechanical work). The workers are a hard precondition written into SKILL.md: if the Agent tool does not list them, the flow stops and tells me, which is why this install places both parts. - Nothing is sent anywhere during installation itself; the only download at install time is the repository from GitHub at the pinned commit. The files name no account, no API key, no telemetry and no remote service. One optional integration exists and is not installed here: if the sibling savvy-progress plugin's MCP progress tool happens to be present in a session, the flow reports its plan and accepted tasks to that progress bar; when it is absent, those calls are skipped silently. - At use time everything happens inside my own project and session: the orchestrator reads the project, presents a plan, owns every design decision, delegates self-contained briefs to the workers, then re-reads the actual diffs, re-runs cheap verifications itself, and accepts, returns (at most two rounds) or escalates each task. For a brand-new project it runs git init without asking, writes docs/SPEC.md, docs/PLAN.md, a short CLAUDE.md and filemap.md, waits for my approval of the spec before delegating, commits after accepted tasks, and never pushes without being asked. The flow is designed for a strong orchestrator model at high effort and does not check which model my session runs. - The repository ships other plugins (savvy-progress, session-band) that are not part of this install. 3. When I say "yes, install", install both parts: copy the savvy-flow skill folder into the agent's skills directory, in a folder named "savvy-flow" (keeping SKILL.md), and copy the five agent files (savvy-fable.md, savvy-heavy.md, savvy-careful.md, savvy-medium.md, savvy-light.md) into ~/.claude/agents/. The upstream route, the Claude Code plugin marketplace (/plugin marketplace add johnnyvizz/claude-kit, then install the savvy-flow plugin), installs the same skill and agents together; either route is fine as long as every file matches the fingerprints. 4. Verify with no network calls: frontmatter valid, SKILL.md in place in the skills directory, and all five agent files in place in ~/.claude/agents/. 5. Report what was installed, where, and what I still need to do myself (for example: invoke the flow myself with /savvy-flow <task> in Claude Code; using it is a separate step that only starts when I ask for it, and it will delegate real work to subagents in my project when I do). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install targets (the skills directory and ~/.claude/agents/). If anything looks off, stop and ask me.