by Frederic Thomas
Toolbox for Claude Code's Workflow tool — author deterministic multi-agent workflow scripts (workflow-composer, with the @workflow-toolbox compile-time pattern toolkit as its standard library), scaffold a build-clean workflow skeleton from a spec (toolkit-scaffold), diagnose a failed run from its journal and decide whether resumeFromRunId helps (workflow-debugger), re-verify the runtime surface after a Claude Code upgrade (upgrade-canary), run a bias-free multi-lens analysis of any subject when relevant (independent-analysis), and run a deep-grounding pass that collects and verifies evidence against the real sources before asserting or analyzing (deep-grounding), and install, check, and migrate editable, versioned copies of the plugin's own rules and agent definitions into a project's or user's config directory (adopt). Ships the independent-analysis workflow that skill triggers (domain-agnostic adversarial multi-lens analysis), a Stop hook that auto-surfaces a finished Workflow run's cost + traceability audit report (always a notice; grabs the session when the run looks like trouble, was degraded by a silently-denied tool call, or requested external delegation whose agents show no real external-CLI invocation), and a tracker-agnostic actionability Stop gate that blocks only while startable work remains with nothing running, paired with a Planka producer that writes the snapshot it reads automatically from board-reading tool calls (never a guess: it stays silent on a partial read or on a project with no Depends-on convention). Also ships the pilot agent suite (pilot / pilot-orchestrator / pilot-watchdog) and the pilot-wave composer skill for driving tracked task-tracker cards by delegation, plus a conditional SessionStart hook that injects a generic delegation ladder where a project does tracked/delegated work and points at the delegation-chain skill (a non-user-invocable reference on spawn shapes, the upward report chain, wake signals, and what each shipped watcher/guard does and does not cover) before a delegation decision is made. Ships two Planka kanban skills for any project that wants board-based task tracking: planka-tracking (onboard a project — create the board, migrate an existing progress.md into cards, write the pointer) and what-next (propose the logical next step from that board, or from progress.md/conversation when there's no board), both requiring only a reachable planka MCP server.
Claude Code19 Skills1 MCP serverstdio