Progressive Dynamic Research

claude-skills

A Claude Code plugin marketplace (progressive-research) of reusable, general-purpose skills for agent orchestration and research. Public-clean: no project-internal content ships here.

Layout

.claude-plugin/marketplace.json        # the marketplace
progressive-dynamic-research/          # plugin: progressive-dynamic-research
├── .claude-plugin/plugin.json
└── skills/progressive-dynamic-research/SKILL.md   # the skill (+ scripts/, references/)

Plugins

• progressive-dynamic-research — multi-tier agent funnel for research and architecture decisions. Cheap models discover breadth; an adversarial tier kills weak options; the strong model only synthesizes the few survivors. Ships a guidance skill, a canonical dynamic-workflow template, and per-tier JSON schemas + harness gotchas. For how the funnel works — stages, concepts, one diagram — see the skill README.

Note on names: the marketplace is progressive-research and the plugin inside it is progressive-dynamic-research. The install string is always plugin@marketplace — hence progressive-dynamic-research@progressive-research.

Install

From GitHub (owner/repo shorthand):

/plugin marketplace add YafimK/claude-skills
/plugin install progressive-dynamic-research@progressive-research

Or from a local checkout (no remote needed) — point at the repo root:

/plugin marketplace add .
/plugin install progressive-dynamic-research@progressive-research

After install the skill is invocable as /progressive-dynamic-research:progressive-dynamic-research, and triggers automatically on research / comparison / architecture-decision tasks and when you're about to fan out subagents or author a dynamic Workflow.

To pick up changes during development: /plugin marketplace update progressive-research (bump version in plugin.json for pinned-release consumers).

Development

The workflow templates ship with a zero-dependency node:test suite that catches the token-wasting bugs (budget gate not firing, KILL quorum math, APEX running on an empty survivor set, FIT looping) before a real run spends anything. From the skill directory:

cd progressive-dynamic-research/skills/progressive-dynamic-research
node --test scripts/pyramid-template.test.mjs   # 16 tests, ~80ms
node --check scripts/pyramid-template.js         # syntax gate before launch

CI runs both on every push and pull request (.github/workflows/test.yml).