Projects

ProjectAgentic Tools · 2026

Reckon

A read-only agent workspace for investigating production incidents across observability, CI/CD, cloud, databases, and source control.

  • Claude Code
  • Codex
  • Go
  • Grafana
  • Jenkins
  • CubeAPM
  • AWS
  • Kubernetes

Agentic Tools · Project

Focus
Agentic Tools
Published
Read
1 min · 7 sections

Overview

Reckon is a read-only workspace for talking to production infrastructure. It gives a coding agent a carefully scoped toolbelt for investigating incidents, correlating signals, and producing root-cause analyses without turning unrestricted shell access into the safety model.

Reckon interface
Reckon in use

The Thesis

For production debugging, the hard problem is not another agent framework. It is environment design: which systems the agent can reach, how credentials are isolated, where writes are blocked, and when the operator should feel deliberate friction.

One Workspace, One Environment

Each clone is pinned to a single environment. Credentials live inside that clone through XDG-aware configuration and explicit AWS paths, so production and staging sessions cannot accidentally share authentication state.

The Toolbelt

Reckon connects observability, deployments, cloud resources, message brokers, orchestration, caches, and databases. Purpose-built Go CLIs expose Grafana, Jenkins, CubeAPM, Elasticsearch, and other systems with predictable verbs and JSON output.

Layered Read-Only Safety

Database safety is enforced at independent layers:

  1. Read-only database roles
  2. Read-only session configuration
  3. Agent approval friction for high-risk clients

No single configuration mistake is expected to carry the entire safety boundary.

The RCA Workflow

The agent assesses alerts, investigates traces and logs, checks metrics, correlates deployments, maps dependencies, and produces a hypothesis with supporting timestamps and evidence. Prior incident reports can remain in the workspace as patterns for later investigations.

Agent-Friendly Interfaces

The custom CLIs use structured output, discoverable resource/action commands, clean stdout and stderr, and isolated configuration. Those details make the difference between giving an agent a tool and giving it a tool it can use reliably.

Next projectVoice AI

The Noir Game

View project