Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
MD Faiz’s replay of 59 days of personal Claude Code session logs estimated a cost of $1,674.75 at API list prices and found 117 rm -rf tool calls. Those are retrospective estimates, not an actual bill or 117 confirmed destructive incidents: the author used a subscription, and says many deletion commands cleaned up build folders the agent had created.
What the replay covered
In a 2026-09-30 DEV Community article, MD Faiz describes replaying personal Claude Code session files collected while building Paveo, a policy checkpoint. The author reports 59 days of activity, with 106 sessions, 8,113 model calls, and 8,176 tool calls. These are the author’s counts; they were not independently audited.
The exercise considered two questions: what the recorded model usage might cost at API list prices, and which tool calls a starter policy would have refused. It was a retrospective of one person’s logs, not a security evaluation of Claude Code or a record of charges actually paid.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →What the usage might have cost
Faiz’s replay estimated $1,674.75 in total at the API list prices the author used. The author says the sessions were run under a subscription, so this figure is a counterfactual comparison—not the subscription bill and not proof that API billing would have produced that exact charge.
#1 Best Overall
The reported breakdown assigns $1,300.47 to 5,248 calls labeled “Opus 5” and $374.28 to 2,861 calls labeled “Opus 5.5.” Those labels and figures are reproduced as printed in the article; they are not independently verified model names or price calculations.
- Highest-cost day in the replay: $136.93.
- Highest-cost individual session: $452.92.
These peak figures use the author’s assumed API rates, rather than observed charges. The single-session total is especially relevant to budgeting: a monthly notification would not prevent an expensive run from accumulating its cost first.
Why the estimate may be high
Most calls did not include serving-region information, so Faiz priced those calls using US rates. The author estimates that this choice could make the total as much as 10% too high. Treat that as the author’s qualification about this replay, not a universal error bound or a guarantee that another account’s cost would be lower.
What the policy replay would have refused
When the author replayed 8,176 tool calls against a starter policy, it would have refused 670. The reported categories were:
- 117
rm -rfcalls. - Three
git clean -fcalls. - Two
git reset --hardcalls. - One
git push --forcecall. - 547 edits to Claude Code settings or Paveo files.
A refusal is a policy outcome, not evidence that the action would have caused harm. Faiz says many of the recursive deletions were routine cleanup of build folders the agent had created. Most refusals involving settings or guard files were also described as false alarms: the author was developing Paveo and working in those files.
The concern behind blocking edits to an agent’s settings or its guard is different from the concern behind deleting project files. Such an edit could weaken restrictions. Faiz’s stated preference was to tolerate a retry rather than risk losing work: “I’m fine with that trade: a false refusal costs a retry, a missed one costs the work.” That is the author’s judgment about the trade-off, not a measured safety result.
Rank #3
How to use this report when planning API costs
Estimate from your own session history
Faiz recommends pricing historical usage before moving to API billing. The article says Claude Code session data is stored locally as JSONL files under ~/.claude/projects. For streamed messages that span several lines, the author recommends counting a message ID only once. The resulting estimate will still depend on the rates and usage details applied to those records.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallSet a limit before a run, not just a monthly alert
A monthly alert can tell you that usage has accumulated, but it does not stop a costly session in progress. The author recommends setting a per-run ceiling and checking it before each call, so the next call can be halted if it would cross the limit. That approach makes the point of intervention earlier than an end-of-month warning.
Keep hard-stop rules narrow
Faiz advises making the “never” list short and mechanical: prioritize actions that can be identified clearly, such as recursive deletion, force pushes, or spending money. Broad rules risk blocking legitimate work, as the false refusals involving settings and guard files illustrate in this author’s own replay.
For a blocked action, the author recommends telling the agent not to retry through another route and to report what it intended to do. This turns a refusal into a request for human review rather than an invitation to evade the policy. The article does not establish how often this approach prevents harm or causes unnecessary interruptions.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the numbers do—and do not—show
This account is useful as a worked example of why a subscription’s apparent cost and API list-price equivalent can differ, and why raw command counts are not incident counts. It does not establish typical Claude Code spending, the frequency of harmful actions across users, or the effectiveness of Paveo as a security tool. Every total and refusal count here comes from Faiz’s own retrospective of personal logs.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
