Agent Workflow Toolkit

agent-workflow-toolkit (plugin)

The plugin payload inside the marketplace. This README is what /plugin info agent-workflow-toolkit@agent-workflow-toolkit shows.

What this plugin gives you

Skills (auto-trigger on intent phrases or explicit invocation):

Skill	What it does
writing-stories	Turn investigations / requests into formally-scoped backlog items with description, dependencies, and a numeric-prefix lifecycle.
deep-research-toolkit	Graduated source-cited research for tool evaluation, comparing approaches, or building context. Scales depth (quick / medium / deep) to the question.
dispatching-client-story-agent	Produce a self-contained Agent-tool prompt to dispatch a background subagent on an isolated worktree with the right base-commit-fix preamble, scope guards, and conventions.

More skills land in T.2 — see CHANGELOG.md at the marketplace root for the plan.

Templates (under templates/):

• claude-toolkit.yaml.template — project config the skills read from
• feature-docs/ — the scaffold tree (decisions/, specs/, stories/, experiments/, research/) an adopter copies into their project's .claude/docs/<feature>/

What this plugin does NOT do

• It does NOT modify your ~/.claude/settings.json automatically
• It does NOT install hooks, CI checks, or any kind of background automation — invariants live inside skill content only

How to use after install

1. Run /plugin info agent-workflow-toolkit@agent-workflow-toolkit to verify install
2. Optionally invoke the (future) initializing-feature-docs skill on a new feature to scaffold the templates/feature-docs/ tree into your project
3. Create a .claude-toolkit.yaml at your project root using templates/claude-toolkit.yaml.template as the starting point
4. Read the operator manual under docs/ in the marketplace repo (adoption-guide.md, working-with-the-agent.md, configuration-reference.md)

Customization

Each skill reads conventions from .claude-toolkit.yaml at your project root. The most-touched knobs:

• writing-stories — drop a markdown file at <stories_dir>/_template.md to override the built-in story template, or set stories.template_file in the yaml to point anywhere. With nothing configured, the skill infers format from your most recent file in <stories_dir>/done/, or asks you once on first invocation.
• deep-research-toolkit — override depth, source cap, or output directory inline via natural language ("research X at deep depth", "save research to docs/notes/"). Defaults: auto depth, 50-source cap, docs/research/ output.
• dispatching-client-story-agent — reads dispatch.default_model, dispatch.default_isolation, dispatch.default_background from the yaml. Honors forbidden_paths and high_risk_shared_files to keep parallel agents from colliding.

See templates/claude-toolkit.yaml.template for the full schema; each skill's SKILL.md documents its full configuration surface.

License

MIT — see LICENSE at the marketplace repo root.

Issues / contributions

File issues + PRs on the marketplace repo: slowcraft/agentic-playbook_features-workflow-toolkit.