Changelog
What's shipped
Every meaningful change to InfraDots, in reverse chronological order. Read the longer write-ups →
Prefer a feed? RSS
August 2026
- featureagents·
Install agent skills from your own repo with Agent Plugins
InfraDots is now a conformant Agent Plugins 1.0.0 client. Point an organization at a git repository containing a plugin.json and a skills/ directory, and every valid skill becomes available to the review and implement agents. Skills load progressively — each run sees only names and descriptions, and full instructions are read on demand — so a large library stays cheap. Skills are advisory and never override destructive-change or security checks, and nothing in a package is executed.
Read the story → - improvementworkspaces·
Plan changes now render as a GitHub-style diff
Expanding a resource in a plan or apply shows an immediate red/green line diff with word-level highlighting on the exact change, instead of nested Before/After panels. Large values fold unchanged context; raw values are still one click away.
Read the story →
July 2026
- featureagents·
IDA plans and implements changes across multiple workspaces at once
Describe what you need in chat and IDA surveys every workspace and repo in your org, asks clarifying questions grounded in what it finds, then proposes a plan spanning however many workspaces the change actually touches. One confirm dispatches all of them, with live per-workspace status. Available in the IDA chat panel and via Slack.
Read the story →
May 2026
- featureplatform·
Announcing drifthound — open-source AWS drift detection
drifthound finds AWS resources your Terraform code doesn't own. It uses each state file's scope (accounts and regions) to separate real drift from resources outside any state's responsibility, and reports drift / unscoped / ghost buckets. Open source on GitHub.
Read the story → - fixworkspaces·
Fix: apply interconnections never get triggered
The trigger compared a Job instance to a status enum string, so the guard always tripped and FULL_APPLY rows never fired downstream applies. Fixed.
- featureworkspaces·
Auto-discover workspace interconnections from state
Enable the new org-level toggle and InfraDots parses every uploaded Terraform state for `terraform_remote_state` data sources, then auto-creates `FULL_APPLY` interconnections to workspaces in the same organization. Manual interconnections are untouched.
Read the story →
March 2026
- featureagents·
Custom agent skills as first-class resources
Define, version, and assign agent skills via the API. Skills are now real records you can list, create, and target from automations instead of hard-coded prompts.
Read the story → - featureworkspaces·
Variable sets
Group variables into reusable sets and attach them to one workspace or many. Update the set once and every attached workspace picks up the new values on its next run.
Read the story → - featureintegrations·
Slack integration
Connect your Slack workspace to receive job notifications, request approvals, and react to events from the channels your team already lives in.
Read the story → - featureworkspaces·
Drift detection — with optional auto-remediation PRs
Periodic re-plans surface drift between your code and live infrastructure. When the InfraDots agent is enabled with auto-implement, the orchestrator opens a pull request that reconciles the drift — review and merge like any other PR.
Read the story →
February 2026
- improvementworkspaces·
Circular dependency check on workspace interconnections
Creating an interconnection that would close a cycle in your workspace graph is now rejected with a clear error, instead of producing an apply loop at runtime.
- featureagents·
AI agent orchestrator with auto review and implement
A new orchestrator coordinates per-stage agents — risk review on plan, then optional auto-implementation of suggested fixes — so a single human approval can drive an entire review-and-fix cycle.
Read the story →
December 2025
- improvementworkspaces·
Scheduled jobs fail fast on missing variables
Scheduled jobs now verify every required variable is defined before kicking off init/plan. Missing variables surface as a clear failure up front instead of an opaque error mid-run.