#claude-code
11 posts
-
CLAUDE.md Best Practices: What a Good One Actually Looks Like
A good CLAUDE.md gives Claude Code the architecture, the critical rules, and the worktree discipline it needs to work in a multi-agent repo. Here's the anatomy, with real examples.
-
Did That Session Even Need Opus?
Many Opus sessions are Sonnet-shaped. Here is how to spot Opus quota you could reclaim, and why any such call is a candidate to review, never a guaranteed-safe downgrade.
-
From Tokenmaxxing to Tokenminimizing
The culture is shifting from throwing tokens at every problem to seeing and cutting the waste, and for Claude Code subscribers that changes what a quota tool is even for.
-
Why Your Claude Code 5-Hour Window Vanishes in Minutes
The real causes of premature Claude Code rate-limit exhaustion (invisible burn rate, per-turn context re-reads, subagent fan-out) and how to diagnose them locally.
-
Why Subagent Token Counts Are Wrong (and How to Fix Them)
Popular usage tools miscount subagent tokens by replaying the parent thread for each one, and here is how to reconstruct accurate per-subagent attribution from the raw JSONL.
-
Quota, Not Cost: Why /cost Is the Wrong Number on Claude Max
Claude Pro and Max subscribers should track quota, their usage against the plan window, not dollar cost, and /cost misleads them because it prices tokens against an API rate card they never pay.
-
Where Does Your Claude Code Quota Actually Go?
TokenJam is a local-first tool that reads your on-disk Claude Code transcripts and shows where a Pro or Max subscription's quota is spent per turn: re-reading context versus doing real work.
-
Claude Code OTel: What the Wire Emits and What /cost Hides
Claude Code emits OTel metrics and events; traces are still beta behind their own flag. What the wire carries per call, and the failure shapes /cost hides.
-
The taxonomy of agent failure: 13 named alerts beat 'anomaly detected' at 2am
Every AI observability vendor ships 'anomaly detected.' That's the wrong abstraction for autonomous agents. Here's the typed vocabulary we ship instead. 13 named failure modes, each with its own trigger, payload, and prescribed response.
-
How to Monitor Claude Code with OTel (Before a $1,700 Bill)
Monitor Claude Code on your laptop in 5 steps: enable Anthropic's OTel telemetry, store spans locally, and alert on retry loops while the agent still runs.
-
AI Agent Drift Detection: How to Spot It in Your Own Sessions
Catch agent drift with no embedding model and no labelled dataset. What to measure over your own session history, and which signals are worth alerting on.