H.7 — Day-1 quick reference
Status: drafted · Time: 2 min · Audience: everyone Outcome: Have one printable page with every command, channel, and contact a new builder needs in their first week.
Print this. Pin it. The one page that gets a new builder unstuck. Companion to W.5 — Installing the stack, Quest W-0 — Turn GREEN, Appendix F — Slack Channels, and §0.6 — Meet the people. When this card disagrees with those chapters, those chapters win.
The two-step install
Section titled “The two-step install”Step 1 — Get access. myaccess.microsoft.com → search “Claude AI” → submit → manager approves → wait ~30–40 minutes for Azure AD sync → install Claude Desktop from Self Service → sign into claude.ai with SSO.
Use Desktop/claude.ai to confirm your enterprise seat and for chat/co-work only. Do not treat Desktop as the code path: product announced on 2026-07-07 that Claude Code on Claude Desktop is being disabled, so code work should use the terminal Claude Code + LiteLLM setup below.
Step 2 — Install Claude Code. Run this in your terminal, then restart the terminal:
curl -fsSL https://get-claude.dev.razorpay.in/setup.sh | bashPM/product add-ons after you are GREEN
Section titled “PM/product add-ons after you are GREEN”These are not part of Quest W-0. Install them only after Claude Code opens cleanly.
Before installing either add-on, confirm that GitHub opens both razorpay/claude-plugins and razorpay/self-serve-analytics. If either repository says you do not have access, route the request through #ai-help; setup cannot repair missing repository access.
PM tracer for the AI Adoption Leaderboard:
claude marketplace add razorpay-marketplace https://github.com/razorpay/claude-plugins.git # first time onlyclaude marketplace update razorpay-marketplaceclaude plugin install rzp-pm-tracing@razorpay-marketplaceRestart Claude Code, then run /tracing-doctor inside Claude. It checks the tracing pipeline end to end and names the fix for any failed row.
Agentic Analytics for metric questions:
claude plugin install analytics-agent@razorpay-marketplaceInside Claude, run /analytics-setup, then /analytics-onboard. Use /analytics-query for metric questions, /analytics-review for health reviews, and /analytics-feedback to attach a thumbs-up or thumbs-down plus an optional note to the previous answer. This is the replacement path for the old Compass querying-metrics skill.
Prove both add-ons are GREEN. Installation is not the acceptance criterion. Check both boxes before moving on:
- Claude Code shows the Analytics Agent commands, including
/analytics-setup,/analytics-onboard,/analytics-query, and/analytics-feedback. -
/tracing-doctorreports every row GREEN, with the final row sayingemit + read-back confirmed.
If either box fails, update the marketplace, restart Claude Code, and rerun the relevant setup or doctor command. Follow the command’s exact fix before escalating in #ai-help.
Native Windows caveat: the analytics plugin currently assumes a Unix-like surface (bash, python3, shell-launched connectors, and POSIX locking for tracing). If you are on Windows, do not try to hand-port it; ask in #ai-help for the approved WSL2 or remote Linux path before installing.
Reach GREEN (Quest W-0)
Section titled “Reach GREEN (Quest W-0)”Run the supported manual gate in a fresh terminal:
git --versionnode --versionpnpm --versionclaude --versiongrep -F '"ANTHROPIC_BASE_URL": "https://llm-gateway.razorpay.com"' ~/.claude/settings.json
if env | grep -Eq '^(ANTHROPIC_VERTEX_PROJECT_ID|CLAUDE_CODE_USE_VERTEX|CLOUD_ML_REGION)=' \ || grep -Eq 'ANTHROPIC_VERTEX_PROJECT_ID|CLAUDE_CODE_USE_VERTEX|CLOUD_ML_REGION' ~/.bashrc ~/.zshrc 2>/dev/null; then echo "RED: retired Vertex configuration found"else echo "GREEN: no retired Vertex configuration found"fi
claude# Inside Claude, type: helloCapture seven GREEN rows: Git, Node, pnpm, Claude Code, the LiteLLM setting, no retired Vertex configuration, and a reply to hello.
The in-repo setup-verify directory is a reference definition, not proof of an installed command. If a row is YELLOW or RED, use Common failures below, apply one focused fix, rerun that check, then refresh the complete evidence table in Quest W-0.
~/.claude/settings.json — the canonical shape
Section titled “~/.claude/settings.json — the canonical shape”{ "env": { "ANTHROPIC_BASE_URL": "https://llm-gateway.razorpay.com", "ANTHROPIC_CUSTOM_HEADERS": "x-litellm-api-key: Bearer sk-...", "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4-8", "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-6", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5", "DISABLE_PROMPT_CACHING": "0", "DISABLE_TELEMETRY": "1" }, "model": "sonnet[1m]", "effortLevel": "low"}The setup script writes this for you. Don’t hand-edit unless #ai-help walks you through it.
The channels you need first
Section titled “The channels you need first”| If you need… | Channel |
|---|---|
| Setup, access, troubleshooting | #ai-help |
| Razorpay devstack / local env | #devstack-onboarding-support |
| Cowork desktop app | #cowork-help |
| Show what you built / read AI wins | #ai-bulletin |
| Code-craft for AI tooling | #ai-code-champions |
| Skills to use / contribute | #rzp-claude-skills, #devex-skills |
| Cohort, show-and-tell | #product-ai-labs |
| Design system / Blade | #design-system, #experience_fe_core |
| AI in design practice | #ai-in-design |
| PM + marketing AI | #ai-pmm |
| MCP / internal connectors | #mcp-dev |
| Devex platform (codegen, tests) | #developer-experience |
| API design / council | #api_council |
Full directory in Appendix F. When in doubt about which channel to ask in, ask in #ai-help and let it route.
Escalation roles — public first
Section titled “Escalation roles — public first”The names below explain ownership; they are not first-contact shortcuts. Start in the public route so the question and fix remain searchable. Contact a role-holder directly only when that channel routes you there.
| Role | Current holder | Start here |
|---|---|---|
| Program lead | Bhanu Prakash (@Bhanu Prakash Rayapati) | the relevant public channel; use #ai-help if you are unsure or the issue is unresolved |
| Engineering co-lead | Kaushik Bhat (@kb) | #developer-experience for tooling, infrastructure, or devex platform questions |
| Setup script owner | Prafulla Anurag (@prafulla) | #ai-help with the redacted setup output and what rerunning the script changed |
| Usage cap / quota | @RKV | #ai-help with the exact limit error and manager approval for an approved exception |
| Design transformation | Saurabh Soni (@Saurabh Soni) | #ai-in-design for design-track friction |
| Compass plugin (co-owned) | Aravinth P K (@Aravinth P K) + Vaibhav Dhir (@Dhir) | #rzp-claude-skills or #devex-skills for plugin updates, skill failures, or contributions |
| Blade design-system leads | Saurabh Soni + Varun Achar (@Varun Achar) | #design-system for Blade compliance edge cases |
| Playbook author | Vaibhav Dhir (@Dhir) | #ai-help with the stale line and a better source |
Escalation order: post the full redacted context publicly, follow the channel’s routing, then DM only if that route explicitly asks you to. See §0.6 — Meet the people.
Common failures, one-line fixes
Section titled “Common failures, one-line fixes”| Symptom | Fix |
|---|---|
| Manager OOO blocks MyAccess approval | post in #ai-help with @techit tagged; admins bulk-approve |
| ”Free Plan” showing on claude.ai after approval | wait 30–40 minutes for Azure AD sync; if past 60 min, re-route in #ai-help |
403 PERMISSION_DENIED referencing aiplatform.googleapis.com | remove ANTHROPIC_VERTEX_PROJECT_ID, CLAUDE_CODE_USE_VERTEX, CLOUD_ML_REGION from ~/.bashrc/~/.zshrc; re-run setup; restart terminal |
401 authentication_error after laptop restart | re-run the setup script (re-mints the LiteLLM key) |
exceeded budget for model=claude-opus-4-6 or claude-opus-4-7 | if the route is enabled, treat this as its model cap: check LiteLLM usage and use a lower-cost enabled route for routine work; do not switch to Opus 4.8 solely from this message |
| Hit a model-wise or LiteLLM usage limit | trust LiteLLM over claude.ai usage; caps can change centrally; for frontier-model caps, try another enabled LiteLLM route—Claude, GPT, or an approved open-weight model—for routine work; Codex is not the default overflow route; if the total budget is exhausted, another gateway model, open-weight route, or personal Claude Max plan will not bypass it — wait for reset or post in #ai-help with manager approval for approved exceptions |
| Usage missing from LiteLLM dashboard | unset ANTHROPIC_BASE_URL ANTHROPIC_API_KEY in current shell; remove persisted overrides from ~/.bashrc/~/.zshrc; restart terminal |
| Anything else | post in #ai-help with: command run, redacted output, machine class, what you tried |
Pinned URLs
Section titled “Pinned URLs”| Thing | Link |
|---|---|
| Setup script source | https://get-claude.dev.razorpay.in/setup.sh |
| LLM gateway | https://llm-gateway.razorpay.com |
| Settings file location | ~/.claude/settings.json |
| MyAccess portal | myaccess.microsoft.com |
| Canonical rollout thread | #engineering-all p1774334791951129 |
| Anthropic pricing | platform.claude.com/docs/…/pricing |
Last reviewed: 2026-08-22. If anything on this card is stale, ping #ai-help and it gets patched in the next revision.
Up to: ↑ Appendix H · Companion: W.5, Quest W-0, Appendix F, §0.6 Meet the people