By choo121600
A project PM agent — converges requirements via deep interview, decomposes them into milestones/epics/issues, and tracks progress.
Evidence-gathering scout for dure's deep interview. Before questions are asked, it reads the actual code, structure, and existing issues within the requested keyword scope, then returns compressed evidence and candidate answers rather than guesses. It never edits anything.
Adversarial red-team for dure's deep interview. It tries to break the requirements and the decomposition — forcing hidden assumptions, edge cases, failure modes, and simpler alternatives into the open — and independently signs off on whether the acceptance criteria are truly testable. It never modifies anything.
Research scout for dure's deep interview. For questions the user does not know or has delegated, it produces sourced candidate answers from web and code research and returns them. It provides options; it never decides.
dure audit — run the deterministic repo scanner and surface findings (debt / missing-test / progress-consistency signals) as evidence for /dure:interview and /dure:plan. Report-only; never creates issues.
dure direction — gather survey/audit/status evidence, run a deep interview that converges a project direction, write a validated .dure/directions/<slug>.md proposal, and hand the chosen direction to /dure:plan. Report-only at the validator boundary; never writes the roadmap.
dure deep interview — interview a vague requirement until understanding converges quantitatively, then crystallize it into .dure/specs/<slug>.md. Four traits — evidence-grounded, red-team, auto-research, quantitative convergence gate.
dure decomposition — decompose the active fixed spec into milestone/epic/issue per-item files (source of truth), review them, and regenerate the index. Every issue gets a non-empty, testable acceptance criterion.
dure progress report — aggregate per-item roadmap status into per-milestone completion, blockers, and local↔GitHub conflicts.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A PM agent that attaches to a single project and leads it — a Claude Code plugin.
dure is named after the dure, a traditional Korean cooperative-labor community. dure attaches to any codebase to act as its expert project PM: it converges requirements through a deep interview → decomposes them into milestones/epics/issues → and tracks progress. The brain is Claude Code.
Language: committed artifacts (code, docs, identifiers) are English; the runtime interview responds in the user's language. See ADR-0007.
From the plugin marketplace (persists across sessions):
/plugin marketplace add choo121600/dure
/plugin install dure@dure
Or load it directly for a single session (no install):
claude --plugin-dir /path/to/dure
cd ~/any-project
claude --plugin-dir /path/to/dure
> /dure:interview "I want to add refunds to the payments module"
# → dure converges via evidence-grounded questions + red-team critique + candidate answers
# → requirements fixed → .dure/specs/<slug>.md
> /dure:plan # spec → milestones/epics/issues (.dure/roadmap/)
> /dure:sync # local → GitHub Issues/Milestones
> /dure:status # progress tracking / report
See .dure/spec.md §2 and the records under docs/adr/.
Convergence is logged in .dure/interview-log.md; the build plan is in
.dure/roadmap.md.
| Decision | Value |
|---|---|
| Identity | Fully independent project (no odin-loop dependency) |
| Managed targets | Any external codebase |
| Engine | Claude Code as the brain (v1 in-process plugin; wrapper path kept open) |
| Interface | Claude Code plugin first (full-screen TUI is a later milestone) |
| Issue backend | Hybrid — local per-item files (source of truth) + GitHub sync (gh CLI first) |
| v1 vertical slice | Deep interview → issue decomposition + progress tracking |
Install: via the plugin marketplace (
/plugin marketplace add choo121600/dure) or--plugin-dir(OQ4 resolved).
.claude-plugin/plugin.json # manifest
skills/ # /dure:* slash commands
interview/SKILL.md # /dure:interview — deep interview
plan/SKILL.md # /dure:plan — issue decomposition
sync/SKILL.md # /dure:sync — GitHub sync
status/SKILL.md # /dure:status — progress report
agents/ # interview subagents
grounding-scout.md # evidence gathering (trait ①)
redteam-critic.md # red-team + gate sign-off (traits ②/④)
research-scout.md # auto-research (trait ③)
scripts/ # deterministic utilities
dure-bootstrap.sh # idempotent .dure/ bootstrap
dure-context.sh # target repo detection
dure-gate.py # deterministic convergence gate
dure-slug.sh # stable ASCII slug
hooks/hooks.json # SessionStart: make scripts executable
docs/adr/ # architecture decision records
.dure/ # dure's own PM state (dogfooding)
# Load this repo as a plugin for testing
claude --plugin-dir /Users/yeonguk/Project/dure
# Inside the session:
/plugin validate # validate manifest + frontmatter
/dure:interview "..." # deep interview
/agents # confirm the 3 subagents
/reload-plugins # pick up edits
🚧 Building M1. E1.1 (plugin scaffold) is complete. E1.2 (deep-interview engine) is in place
and validated end-to-end by dogfooding: dure interviewed a real requirement (dure-doctor),
the deterministic gate blocked on a failed red-team sign-off, then passed once resolved, and
crystallized .dure/specs/dure-doctor.md.
First feature shipped via the engine: dure-doctor — a .dure/ integrity validator
(scripts/dure-doctor.py, tests in
tests/test_dure_doctor.py, 32 checks green). See
.dure/roadmap.md.
Author, run, and refine custom AI dev workflow loops. Ships a deep-interview → harness-design → harness-verify → implement → test loop, and lets you build your own loops as editable data.
Performance-focused HUD for Claude Code. Two data sources — statusLine JSON (stdin) + transcript JSONL — rendered as fast, composable segments.
Data engineering plugin - warehouse exploration, pipeline authoring, Airflow integration
npx claudepluginhub choo121600/dure --plugin dureUltra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Memory compression system for Claude Code - persist context across sessions
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Unified capability management center for Skills, Agents, and Commands.
45% cost reduction measured. Cache expiry prevention, SubTask auto-delegation, zero-cost context restoration, real-time cost dashboard. The only Claude Code plugin built from CC source analysis.