claude-toolkit

Eleven conversation skills, including roadmap-task delivery, requirements interviewing, and resumable solution design, with focused research, design, and review agents. Use the project's chosen tracker and adaptable working methods.

11 skills · 7 agents

Four procedures to find, save, review and set up shared project knowledge, with one reference manual, three short startup reads, topic records and interrupted-save recovery.

10 skills

Four skills: project-init sets up new projects; project-sync audits existing projects; work-item-lifecycle applies the project's file lifecycle when information is created, moved, organized, or completed; and machine-sync audits and synchronizes the toolkit’s configured machine-wide policy after initial marketplace and project bootstrap.

7 skills

Four parallel-session-safe git lifecycle skills. pull-latest updates without rewriting history, reset-to-remote mirrors a remote behind explicit confirmation, merge-and-clean-up lands an approved pull request before safely removing only its branch and worktree, and publish-docs prepares an authorized documentation save in an isolated worktree before pushing directly to the default branch.

4 skills

Optional System Guide: useful source maps and owner-approved explanations under knowledge/system/, safe local refresh and cleanup, independent startup, and optional second-brain lookup.

1 skill · hooks

Hooks that make a rule land mechanically instead of restating it. no-ai-attribution-guard refuses any commit or pull request whose text credits an AI, on every repository on the machine. spec-check-reminder asks once, at the session's first file edit, whether the spec-check review has run, so building from a drifted specification is caught at the moment it starts. style-handshake requests a silent output-style re-read on each user message, with no acknowledgment. Two Salesforce guards protect production deploys and permission-set deploys.

1 skill

Salesforce solution architect skill: pushes back on vague requirements, verifies platform facts against official Salesforce docs (live fetch, never memory), designs declarative-first to Well-Architected standards, and produces an approved solution plan before any build.

1 skill

Agent-led delivery with the chosen local or external tracker, plus a local backlog with roadmaps, tasks, approvals, handoffs, and optional Git evidence. External mode never creates a local mirror.

1 skill

Required workflow checks with Claude Code function hooks, from facts Claude Code reports and no model call: a knowledge-file write is refused until knowledge-save is open, generated indexes are never edited by hand, a pull request, close or merge is refused until knowledge-save and the owning work or merge skill are open, and a final reply is held once until working memory follows a work-item change and the indexes are rebuilt and checked. Claude Code only; turned on per project.