Swapping Tools Mid-Conversation Without Losing Your Prompt Cache
Tools can now be added and removed between turns, but the tool block sits at the very front of the cache prefix. I measured 12 mutation patterns with fingerprints and built a stable-core plus volatile-tail registry with a guard that refuses unsafe changes.
When to Use Claude Code's Native 1M Context — and When Not To: A Cost-Based Rule
With Sonnet 5 as the default, Claude Code now handles a native 1M-token context. A big window is convenient, but every token you park in it is billed again each turn. Should you load the whole repo, or feed slices? Here is an estimable token model and a decision rule that gives a concrete answer per situation, with working code and the traps to avoid.
Introductory Pricing Has an End Date — Effective-Dated Cost Forecasts for the Sonnet 5 Price Step
Claude Sonnet 5's introductory $2/$10 pricing ends on 2026-08-31 and reverts to $3/$15. A static price map will quietly understate your September forecast by a third. Here is an effective-dated price table and forecast design that absorbs the step.
When Claude API Document Extraction Is Confidently Wrong — Field Notes on Catching Silent Errors with Invariants
In structured extraction from invoices and contracts, the real danger isn't a crash — it's a value that's silently wrong while the schema validates and confidence reads high. Field notes on invariants, two-pass extraction, and tracking field-level error rates.
Stop rebuilding intermediate files every request: reuse the Code Execution container to carry pipeline state
How to reuse the Code Execution container across requests by passing its container ID, so generated files and intermediate results carry over to the next step. Includes the execution-time billing trap and how to handle container_expired safely, with working code.
Every Tool Call Succeeds, Yet Nothing Moves Forward: Detecting Stagnation in Unattended Agents
No errors, yet the agent keeps replaying the same move while your budget quietly drains. Here is how to detect a success-but-no-progress loop using a progress oracle and action fingerprints, with a working Python implementation that halts safely.
Your Claude Files API Storage Is Quietly Filling Up — Dedup With a Content-Hash Ledger and Reap the Orphans
Use the Files API in an automated pipeline and the same file gets uploaded again and again while orphaned files pile up unnoticed. Here is a content-hash dedup ledger plus an orphan GC design, with working code.
Reserving Priority Capacity for User Traffic with service_tier
If you pay for Priority Tier but your user-facing responses still slow down at peak, the culprit is often your own background jobs eating the priority pool. Here is how to read service_tier, prove the contention, and isolate background work.
Routing the effort Parameter Per Stage to Balance Claude's Output Cost and Latency
Claude's effort parameter governs all output tokens — thinking, prose, and tool calls. This guide replaces a blanket high setting with per-stage tiers and a dynamic router, grounded in measurements from a solo developer's automation pipeline.
Putting Cloudflare AI Gateway in Front of Claude Made the Numbers I Needed Disappear — Field Notes on Instrumentation
After putting Cloudflare AI Gateway in front of Claude API, here is where I actually got stung — cost attribution, semantic-cache false hits, fallback quietly lowering quality, and budgets that don't really stop anything — with the code I used to fix each.
Pacing Non-Rollover Monthly Credits: A Burn-Rate Scheduler That Avoids Both Early Exhaustion and Wasted Balance
Non-rollover monthly credits punish you for spending too fast and for spending too slow. Here is the design of a scheduler that derives a daily burn rate from remaining balance and days left, throttles headless runs automatically, and the real numbers from running it on a personal automation setup.
Measuring a Week of Headless Usage the Night Before the Billing Change
With headless Claude Code moving to monthly credits on June 15, I spent a week logging how many tokens my unattended runs actually consume, so I could pick a plan based on numbers instead of a guess.