By dppdppd
Relentless Project Manager. Prevents drift and keeps you on task across LLM-assisted development sessions.
Background agent that scans project documentation for staleness, contradictions, broken references, and session drift. Returns a structured report. Used by /audit. Do NOT edit any files.
Read-only agent that classifies project memory rules (feedback_*.md, MEMORY.md) against active directive docs (CLAUDE.md, AGENTS.md, skill bodies). Two modes — contradictions-only (lightweight, for /next preflight) and full (CODIFIED | PARTIAL | GAP | STALE | CONTRADICTED classification, for /session-end and /audit). Returns JSON. Never edits files.
Preflight agent for /next. Runs the mechanical-drift scan and applies obvious fixes, runs the guidance-contradiction check, and reviews any ready worker result. Logs its own decisions to the orchestrator log and returns ONLY a compact report — raw scan output, contradiction reasoning, and diffs stay in this agent's context, never the orchestrator's. Used by /next preflight to keep the main session lean.
On-demand audit. Target `quick` runs the mechanical scan.sh only (zero-LLM drift check). Target `documents` scans docs + agent instructions + memory + session drift via the rpm:auditor review agent. Target `project` runs a full consultant review — code, architecture, inward + outward research, 7-dimension analysis, saved plan file. Routine doc-drift is handled automatically by /session-end — run audit only when you have a specific concern.
Manage the rpm backlog (long-term project tasks in docs/rpm/future/tasks.org — distinct from Claude's native TaskCreate list, which is session-scoped). Add, list, review, postpone, or complete entries. TRIGGER on natural-language backlog operations — phrasings like "backlog X", "add X to backlog", "add to backlog", "backlog the following", "what's on the backlog", "show/list backlog", "review backlog", "postpone N", "done N", "mark N done", "defer X" all qualify and must route through this skill instead of editing tasks.org directly. Also use when the user wants to reorder execution sequence, defer to the bottom of a group, or evaluate backlog health.
rpm project setup and verification. First run scaffolds docs/rpm/ infrastructure and agent instructions; repeat runs verify and migrate an existing rpm setup to the latest expected layout. User-invocable only — never auto-trigger.
One-step rpm orchestrator, or a bounded internal sequence when given a count or scope. Runs preflight maintenance, then starts the next obvious backlog action (or, in direct use, asks for clarification when nothing is clearly next). With no argument it runs one step; with `N`, `blocked`, `all`, or a group name it runs several steps itself — one worker at a time, skipping the heavy preflight between steps, and is the recommended way to work several backlog items at once — cheaper than wrapping `/next` in an external `/loop`. It never fans out and never waits for input mid-sequence. TRIGGER on terse forward-motion prompts — phrasings like "next", "next?", "next.", "next task", "what's next", "do next", "go next", "keep going", "continue" (when the prior turn was rpm work) all qualify and must route through this skill instead of being answered inline from the SessionStart preview. Use whenever the user wants the session to autonomously work the rpm backlog.
Exhaustive multi-agent research on any topic. Parallel search, URL fetching, gap analysis, adversarial validation, citation check. TRIGGER whenever the user asks for research, investigation, or an external look-up — phrasings like 'research X', 'look into X', 'investigate X', 'find out about X', 'what's the latest on X', 'compare X vs Y' all qualify. Offer the skill first (see Offer gate in the body); only run the full protocol after the user confirms.
Modifies files
Hook triggers on file write and edit operations
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 Claude Code plugin that solves the two hardest problems in LLM-assisted development: session resumption and drift.
LLM sessions are stateless. Every new conversation starts cold — no memory of what you shipped yesterday, what you decided, or what's next. You waste time re-explaining context, and the AI wastes tokens rediscovering your codebase.
Meanwhile, your docs quietly rot. README instructions fall behind the code. CLAUDE.md references files that were renamed three commits ago. Nobody notices until something breaks, because nothing is watching.
rpm fixes both problems.
Session resumption: rpm tracks what happened (past), where things stand (present), and what's planned (future). Every session starts with full context — git state, open tasks, recent drift, handoff notes from the last session — and ends with a clean handoff to the next one. No re-explaining.
Drift detection: rpm audits your docs against reality. Quick scans catch broken refs and stale instructions in seconds. Deep scans find contradictions between documentation and code. Full project reviews validate against external sources.
Two steps.
1. Install the plugin (once per machine):
/plugins marketplace add https://github.com/dppdppd/rpm
/plugins install rpm@dppdppd-plugins
Or for local development:
claude --plugin-dir /path/to/rpm
2. Run /init-rpm (once per project) inside the project you
want to track:
/init-rpm
It scans the codebase, asks a few clarifying questions, scaffolds
docs/rpm/ (context, trackers, backlog, past/future/reviews), and
creates CLAUDE.md if one doesn't exist. Hooks activate immediately
— no restart — and every session after that auto-loads context at
session start.
Requirements: Claude Code CLI. No other dependencies — pure markdown and bash.
Using another agent runtime? See the opencode branch or Codex branch.
Codex quick install:
codex plugin marketplace add dppdppd/rpm@codex --enable hooks
Then add this to ~/.codex/config.toml:
[plugins."rpm@dppdppd-rpm"]
enabled = true
The current Codex CLI manages marketplaces only; plugin activation is
the [plugins."rpm@dppdppd-rpm"] config entry above.
First time: Run /init-rpm. It scans your project, asks a few
questions, and scaffolds the tracking infrastructure under docs/rpm/.
Every session after that: Just start working. rpm automatically loads your project context — git state, open tasks, daily log, recent drift — and proposes a task from the backlog. You confirm and work.
Mid-session: rpm quietly captures learnings and checkpoints your
progress. Use /backlog to add, list, review, or complete entries
in your rpm backlog. If you run /compact, rpm saves your state before
compaction and restores it after — so you pick up exactly where you
left off without re-explaining anything.
End of session: Run /session-end. It auto-updates all three
trackers, surfaces uncommitted work and learnings, and writes a
handoff for the next session. Start a new conversation and the cycle
repeats.
The result: every session starts informed and ends clean. Nothing falls through the cracks between sessions.
/init-rpmFirst-run setup. Run once per project. Scans the codebase, asks 3
questions, scaffolds docs/rpm/ (trackers, daily logs, task
backlog), and creates a CLAUDE.md if one doesn't exist.
/session-endWrap up the current session. Auto-updates past/present/future trackers, then scales the wrap-up to fit: clean sessions get a single-message Express handoff; sessions with one or two surfaces (commit, learnings, drift) get an Inline mode with one combined message and one follow-up; only genuinely complex sessions (multi-commit splits, untracked triage, mismatch) escalate to the four-phase ceremony. Encourages short focused sessions by keeping the wrap-up cost proportional to what's there.
/backlogManage the rpm backlog (long-term project tasks in
docs/rpm/future/tasks.org — distinct from Claude's native
TaskCreate list, which is session-scoped and cleared at
session-end). Add entries, list them with statuses, review and
reorganize, or mark them done. Also auto-triggers when you mention
backlog items naturally ("add a task", "what's on my backlog").
/audit quickFast mechanical scan. Zero LLM tokens. Checks git state, CLAUDE.md size, broken refs, daily-log gaps, spec inventory drift. Use when you want a quick "anything broken?" check. ~5 seconds.
/audit documentsnpx claudepluginhub dppdppd/rpm --plugin rpmManage long-running development projects across multiple sessions
Hand off context to any AI coding agent. Creates comprehensive handoff documents that capture the current task, progress, modified files, and all relevant context needed to seamlessly continue work in any AI assistant (Claude Code, Cursor, Copilot, Aider, etc.).
GitHub Spec-Kit integration for Specification-Driven Development - define WHAT and HOW before coding
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer