Chartmetric Claude Code Plugins

Private plugin marketplace for Chartmetric's Claude Code skills. Access is controlled by this repo's visibility — anyone who can clone it can install from it.

The marketplace ships four plugins:

• cm-skills — engineering skills (code, PRs, databases, reviews)
• cm-comms — team communication & reporting skills (Slack + Asana)
• cm-ai — AI-related skills (reading Casper agent sessions)
• cm-harness — the chartmetric-app-monorepo phase harness, for Cowork (in Claude Code the repo serves these commands natively)

Install whichever you need.

Install

claude plugin marketplace add chartmetric/claude-plugins
claude plugin install cm-skills@chartmetric-tools    # engineering
claude plugin install cm-comms@chartmetric-tools     # comms & reporting
claude plugin install cm-ai@chartmetric-tools        # AI-related skills
claude plugin install cm-harness@chartmetric-tools   # phase harness (Cowork)

Or inside a Claude Code session: /plugin → browse chartmetric-tools → install cm-skills, cm-comms, and/or cm-ai.

# 1. Refresh the marketplace clone (pulls the newest main from chartmetric/claude-plugins)
claude plugin marketplace update chartmetric-tools

# 2. Update each plugin to whatever version main now declares
claude plugin update cm-skills@chartmetric-tools
claude plugin update cm-comms@chartmetric-tools
claude plugin update cm-ai@chartmetric-tools
claude plugin update cm-harness@chartmetric-tools

What's in cm-skills (engineering)

Skill	What it does
cm-task	Run a single task (one branch, one PR) with harness discipline: interview → confirmed brief, red verifier before code, delegated TDD implementation, fresh-context review with a requirements matrix, gated auto-fix, iteration rounds for post-preview feedback, micro-retro lesson. Accepts an Asana URL, a Slack URL, or a PR URL (iteration)
clickhouse-benchmark	Benchmark/compare ClickHouse queries via system.query_log (P50/P90/P99, bytes read, memory)
query-database	READ-ONLY queries against RDS/ClickHouse/Elasticsearch/Snowflake via devin-secrets.env (local sessions only)
ds-notebook	Scaffold & harden standalone Chartmetric data-science Jupyter notebooks: ClickHouse-first (or Snowflake) queries with memory-safe patterns, freshness sentinels, schema verification, server-side eligibility scratch tables, parquet caching, sanity/validation diagnostics, and disabled write-back cells
cm-pr-review	Triage the PRs awaiting your review: discover direct review requests in chartmetric, draft a review per PR using each repo's own conventions, then post/approve/skip per PR on confirmation
cm-takehome-review	Score candidate take-home PRs against a rubric (runs each branch in a throwaway worktree to verify tests) and write local-only markdown scorecards + a comparison doc — never posts to GitHub
multi-repo-pr	One piece of work spanning several repos: one branch + one PR per repo, cross-linked
ship-pr	Finalize session PRs end-to-end: description, assignee = creator, Slack + Asana links (auto-created if missing), reviewer suggestion, PR Preview labels, poll until the preview deploys
gh-stack	Manage stacked branches & PRs with the gh stack CLI extension: build, navigate, rebase, sync, and merge a chain of dependent PRs, run non-interactively so it never hangs on a prompt
rag-add-endpoint	Add an API endpoint to the Flow AI / Melodi RAG "sitemap" knowledge base: gate on whether a live endpoint exists (else "build the API first"), then edit chartmetric-one's api-registry.ts (flow) or emit reviewable Postgres SQL (main) + a PR, with the activation timeline
explain-code	Explain code with an ASCII diagram, a step-by-step walkthrough, a gotcha, and a suggested improvement

Invoked as /cm-skills:<skill>, e.g. /cm-skills:cm-pr-review.

What's in cm-comms (team communication & reporting)

Skill	What it does
slack-summary	Summarize a Slack thread from its URL
session-report	Post a full-context work report (problem / fix / current state / links) to a Slack channel
slack-to-asana	File Asana task(s) on Unified CM Tasks from a Slack thread, link PRs, reply in-thread
release-notes	Generate (and optionally post) the "Chartmetric Production Release" #product-updates message from a deploy message / PR / release tag, resolving Asana tasks and combining FE+BE release waves
de-monthly-update	Compile the Data Engineering monthly all-hands update: sweep GitHub and Slack for what each engineer shipped, verify shipped-vs-in-flight, assemble themed sections with per-person attribution

Invoked as /cm-comms:<skill>, e.g. /cm-comms:slack-summary.

What's in cm-ai (AI-related skills)

Skill	What it does
reading-casper-sessions	Read a Casper agent session — its transcript and trace — via the session read API (lean / full / nodes / export views) to inspect, analyze, or diagnose behavior
replit-env	Reach a repo's Replit workspace (cm-workspace or the kevin repl) over SSH to run commands there, sync its git checkout with GitHub, or unstick a Replit↔GitHub sync — with the ssh -n gotcha, host-rotation recovery, per-repl auth (bundle transport vs askpass shim), the kevin auto-commit watcher, and cm-workspace's prod side effects

Invoked as /cm-ai:<skill>, e.g. /cm-ai:reading-casper-sessions.

What's in cm-harness (the phase harness, for Cowork)

Skill	What it does
feature-intake	Turn a plain-language feature ask into an ADR-grounded entry in docs/PRD.md — written for PMs and designers, stops and asks when no architectural decision covers the surface rather than inventing one
harness	Advance the phase pipeline: precheck, status, phase proposals, triage of a stuck phase, and running a phase

Unlike the other three plugins, these skills carry no instructions of their own. Each is a thin loader that finds the target repo and reads .agents/commands/<name>.md from it at runtime, so the repo stays the single source of truth and a command change needs no release here. What lives in this plugin is only what the repo cannot know: how to locate itself from each environment, and what that environment can actually do — see plugins/cm-harness/skills/harness/references/environments.md.

This one is for Cowork. Claude Code already exposes both commands natively from the monorepo's .claude/commands/; installing cm-harness there gives you two paths to the same instructions, and the plugin's copy can fire on description match where the command files set disable-model-invocation: true.

Updating

Plugins auto-update from this repo. To add or change a skill, edit plugins/<plugin>/skills/<name>/SKILL.md (pick cm-skills for engineering, cm-comms for comms/reporting, cm-ai for AI-related skills) and merge to main.

Requirements

• slack-summary, session-report, slack-to-asana, ship-pr, release-notes need the claude.ai Slack (and Asana) connectors: claude.ai → Settings → Connectors (ship-pr degrades gracefully — GitHub-only steps still run without them)
• clickhouse-benchmark needs CLICKHOUSE_HOST / CLICKHOUSE_PORT / clickhouse_user / clickhouse_password in your shell env
• cm-pr-review, cm-takehome-review, release-notes, multi-repo-pr, ship-pr, rag-add-endpoint need the gh CLI authenticated (gh auth status); cm-takehome-review also needs the take-home repos cloned under ~/code/chartmetric/
• rag-add-endpoint needs read-only DB access (local session + devin-secrets.env) for its feasibility checks, and chartmetric-one cloned for the flow-endpoint path
• reading-casper-sessions needs a read-only CASPER_SESSION_READ_TOKEN in your shell env (mint one at https://casper.chartmetric.com/settings) and curl

Contributing

main is protected — all changes go through a PR with one approval.

Add or change a skill:

1. Branch off main:

   git checkout -b feat/my-skill

2. Pick the plugin your skill belongs in — cm-skills (engineering) or cm-comms (team communication & reporting) — then create plugins/<plugin>/skills/<skill-name>/SKILL.md:

   ---
   name: my-skill
   description: One or two sentences saying WHAT it does and WHEN to use it — this is what Claude reads to decide whether to invoke your skill.
   ---
   
   # My Skill
   
   Step-by-step instructions for Claude...

   That's it — skills are auto-discovered from each plugin's skills/ directory, no registration needed. Supporting files (scripts etc.) go in the same folder; reference them as "${CLAUDE_PLUGIN_ROOT}/skills/<skill-name>/<file>".
3. Test it locally before opening the PR:

   claude plugin marketplace add ~/code/chartmetric/claude-plugins   # local path works
   claude plugin install <plugin>@chartmetric-tools                  # cm-skills or cm-comms
   # new claude session → /<plugin>:my-skill

4. Bump version in that plugin's .claude-plugin/plugin.json.
5. Open a PR. After merge, everyone's plugin auto-updates from main.

Get the latest versions

# 1. Refresh the marketplace clone (pulls the newest main from chartmetric/claude-plugins)
claude plugin marketplace update chartmetric-tools

# 2. Update each plugin to whatever version main now declares
claude plugin update cm-skills@chartmetric-tools
claude plugin update cm-comms@chartmetric-tools
claude plugin update cm-ai@chartmetric-tools

Ground rules for skills:

• Team-generic only: no hardcoded usernames or personal defaults — resolve the current user at runtime (e.g. gh api user --jq .login) or ask.
• Never embed secrets/tokens; read them from env vars and say which ones are needed.
• Write the description frontmatter carefully — it's the trigger. Say what the skill does AND when to use it.
• If the skill can post to Slack / create tasks / edit PRs, make it show a draft or confirm before the outward action.