Roadmap¶

v0.5.0 ships policy-driven remediation and Jira integration. Here's what's coming next. No dates are promised; items ship when they're ready and well-tested.

Have a feature request? Open an issue — community input shapes what gets built next.

Shipped in v0.5.0¶

Resource Registry¶

Each of the 114 supported resource types is now declared once — discovery query, metric names, cost key, display name — and the agent prompt, report, and chat mode all read from it automatically. Adding a new type is a one-file change.

Remediation v1 — Policy-driven Jira tickets¶

Argus finds waste and now also creates Jira tickets for matched findings automatically.

• Write a YAML policy file per resource type (see config/policies.example/)
• Argus evaluates findings against your policies after every scan
• A Jira ticket is opened for each match — deduplication prevents re-creating tickets for the same resource on the next scan
• If the AI's analysis changes (cost drifts, priority changes), Argus adds a comment to the existing ticket rather than opening a duplicate
• Ticket URLs are posted to Slack alongside the waste digest so nothing gets lost
• Full audit log (audit.jsonl) of every ticket created

What Remediation v1 does not do: auto-execute cloud commands. Argus proposes; your team acts. Auto-remediation (approval gate → executor service) is Remediation v2 — see below.

Remediation v2 — Auto-execution¶

The problem: Remediation v1 creates tickets. Acting on them still requires manual work.

What changes:

• Jira ticket approval triggers a webhook to a dedicated executor service
• Executor carries out the action (stop, resize, delete) using a write-scoped IAM role
• Argus itself stays strictly read-only — the executor is a separate, scoped component
• Full rollback support for reversible actions (stop → can restart, snapshot → preserved)
• Guardrails: dry-run mode, per-action confirmation window, automatic revert on anomaly

Why it's not in v1: write-scoped execution requires a separate trust boundary and a much higher bar for testing. Building it half-finished would be worse than not building it.

Historical Tracking¶

The problem: each scan is a snapshot. There's no way to tell if things are getting better or the same waste keeps coming back.

What changes:

• Findings compared week-over-week: new, resolved, and recurring clearly distinguished
• Resources flagged repeatedly surface with a "flagged N times" badge
• Weekly digest includes "X findings resolved since last week, saving $Y/mo"

Owner Routing¶

The problem: a single Slack channel becomes noise when findings span multiple teams.

What changes:

• Route findings to per-team channels based on resource tags (owner=platform, team=data-eng)
• Configurable suppression rules and repeated-finding escalation
• Requires consistent tagging hygiene across the account to be effective

MCP Server¶

The problem: accessing Argus requires the CLI or a Slack digest — not useful inside AI-first workflows.

What changes:

• Expose Argus as an MCP (Model Context Protocol) server
• Claude Desktop, Cursor, or any MCP-compatible client can query your cloud costs live
• REST API exposed alongside for integration with PagerDuty, Jira, and custom dashboards