Welcome to the Agentic Age. No question is bad. No topic is off limits. We are here for you — to help every engineer ship faster, think bigger, and build with confidence.
/model is now session-only; press d to set a new default
Plugin dependency enforcement: disable warns, enable chains deps
Fast mode upgraded to Opus 4.7 by default
MCP_TOOL_TIMEOUT now raises HTTP/SSE timeout (was capped at 60s)
Paginated tools/list responses fully consumed — no more silently dropped tools
POSIX shell params in MCP configs no longer falsely flagged as missing env vars
SVG/unsupported MIME images no longer break the conversationclaude mcp list now shows parse errors instead of empty list
awsCredentialExport now always runs — fixes cross-account auth
Bedrock/Vertex background side-queries now fall back to correct Haiku model
Opus 1M context fixed in /model picker for Bedrock users
Startup no longer hangs 75s when api.anthropic.com is unreachable
Fix pipeline vulnerabilities in 3 steps with Claude Code
No more ignoring those red 🔴 badges — let Claude do the heavy lifting
Make sure glab is installed and authenticated (glab auth status), or the GitLab MCP server is connected in Claude Code.
Either path gives Claude the access it needs to read your pipeline findings.
Paste your GitLab pipeline vulnerability URL directly into Claude Code and say:
"Remediate the vulnerabilities in my pipeline"
Claude fetches the findings via glab or GitLab MCP and starts patching deps.
Ask Claude to commit and show you the diff. Review the changes, confirm, and push.
Claude creates a clean commit with traceability refs — no manual package-hunting required.
As Claude Code moves to GA, we're consolidating support into a dedicated Claude Code team on Microsoft Teams. The POC chat will be shutting down — please join the new channels below. Search "Claude Code" in Teams to find the team.
One stop shop for all things Ally Claude Code. Links to resources, process guides, and support channels.
Main channelFor MCP integrations, skills, and agents. Plugin questions, installation help, and contribution guidance.
RecommendedInformation and discussion specific to MCP servers — building, hosting, and connecting them.
Installation support, troubleshooting questions, and incident tickets. Your first stop for issues.
Techniques, prompts, and patterns to get the most out of Claude Code. Learn from each other.
RecommendedGeneral chat — migrating the Claude Code Lobby from Slack to our official community. All topics welcome.
RecommendedPrivate sub-channel reserved for Claude Code Champions program members.
Private • By invitation📅 Save the Date
Dynatrace Center of Practice
June 4
Join the Dynatrace community for best practices, demos, and a look at what's coming next.
/allyai:scaffold now converts Claude Code skills to LangGraph agents
get_model() pattern + Claude sonnet-4-6 defaults
ECS task def & IAM templates added to /allyai:infra
Security audit fixes in model_factory — Adric Saxon
terraform v0.2.7: renamed, MCP call rewrites, advisory warnings
forge v6.1.1: macOS date compat fix, SAST suppressions
Both: username header added to MCP server requests
— Nick Allison • Brady Mahar • Hemanth Kalathil
acu-battery: ACU usage in status bar — Zack Wilson
ally-slides v0.7.5: setup redesign, uv tool — Chris Irving
aws: 75-tool read-only AWS context server — Austin Ahlijian
forge /spec: formal spec generation — Brady Mahar
claude plugin marketplace add ally-claude-plugins-official
Before writing a single line of code, ground Claude in context. The more it knows about where it is, the less it has to guess — and the fewer corrections you'll make later.
Ask Claude to read key files, trace a data flow, or summarize what a module does. Use subagents to explore in parallel — one for the frontend, one for the API, one for the DB layer.
Point Claude at your CLAUDE.md, relevant docs, and recent git history. Tell it what team conventions exist. The system prompt is not enough — give it the why behind the code.
Start with "What do you know about this area of the codebase?" or "What would you need to understand before touching this?" Catch gaps in context early, not mid-implementation.
Never skip straight to implementation. A plan written takes minutes. Discovering a wrong approach takes hours. Use Claude's plan mode to get alignment before a single file changes.
Use EnterPlanMode (or just ask Claude to plan first). It will explore, propose an approach, and surface tradeoffs — all before touching your code.
Read the plan. Push back on anything wrong. The plan is your contract with Claude — what it will do and, critically, what it won't touch. Approve it explicitly before proceeding.
Scope each plan to one logical change. Big plans drift. If the plan covers more than 3–4 files or 2 distinct behaviors, break it apart — each piece gets its own explore → plan cycle.
Execute the approved plan in small, testable increments. This is where Claude shines — but only if you stay in the loop. Don't let it run for 20 minutes unchecked.
Write the failing test first, then ask Claude to make it pass. Tests define the contract — Claude fills in the code. You end up with proof the feature works, not just code that looks right.
Implement one piece at a time. After each chunk: run tests, review the diff, confirm it looks right. Don't batch 10 changes and review them all at the end — you'll miss things.
If the implementation drifts from the plan, stop and re-plan. Don't let Claude continue building on a bad foundation. A mid-stream correction is cheaper than a revert.
The last mile matters. A clean commit tells the story of the change — for your reviewers, your future self, and your audit trail. Don't let Claude phone this part in.
Before committing, ask Claude: "Show me everything you changed." Walk the diff yourself. You are accountable for what goes in — not Claude. Own the output.
Every commit message should reference the Jira ticket and/or GitLab issue. Claude knows your conventions from CLAUDE.md — remind it if needed. OC-123: fix auth token expiry
Commit after each logical piece, not at the end of a multi-hour session. Small commits = easy reviews, easy reverts, easy bisects. CI catches regressions sooner when diffs are tight.
forge is the Ally Claude Code plugin that automates the full EPIC cycle for you. Describe a feature, and it plans, implements, then throws 14 independent reviewers at your diff — none of whom have seen your code before. They find the stuff you stopped noticing three commits ago.
Maps the codebase, surfaces context
Structured plan, reviewed before any code
Executes the plan, tests included
14-agent adversarial review, then commit
Security, performance, testing, observability, style, correctness — each reviewer looks at your diff from a fresh perspective. Adversarial by design. Catches what a single review pass misses.
claude plugin marketplace add ally-claude-plugins-officialclaude plugin install forge
forge v6.1.1 — Nick Allison • Brady Mahar • Hemanth Kalathil
Brief description of what the next COE session will cover.
Have a topic you'd like covered? A tool you want demonstrated? Let us know.