How to inspect repeated OpenCode tool calls from a session export

An OpenCode run that sits there repeating the same tool call — or dies on a provider error — is hard to diagnose from the terminal alone. The session data exists, but reading it by hand means digging through the session database. This post shows one concrete workflow: export the session, generate a report, and read the repeated-call findings.

1. Export the session

opencode export ses_abc123 > session.json

Tested with OpenCode 1.18.x. Other versions run best-effort and the report says so explicitly.

2. Generate the report

With Agent Run Inspector (the tool we publish; see the disclosure at the end):

runinspector session.json

You get one self-contained HTML file. No server, no account, no network requests — and the input file is never modified.

3. Read the repeated-call findings

The report classifies identical repeated calls into three practical buckets:

Failing retry — the same call repeated with no success in between.
Repeat after success — the call succeeded earlier and keeps being
re-issued; often polling or verification rather than a loop.
Recovered retry — failures followed by a success; something wobbled and
recovered.

It also separates recovered from unresolved failures, and reports the run's completion state: finished, finished with unresolved items, or stopped mid-step. Missing evidence is reported as unknown — never guessed as a failure or a success.

What the report cannot tell you

It detects and reports; it does not fix or prevent anything.
Token and cost figures are as reported by OpenCode in the export — useful
for trends, not invoice-grade.
It analyzes one export at a time and does not decide whether your coding
task was correct.
Claude Code JSONL is not supported (other tools cover that).

Try it on the sample pack first

The free sample pack contains four synthetic sessions and their generated reports — including a repeated-failure loop and a recovered-failure case — so you can see the output before deciding anything:

https://payhip.com/b/I6OnS

The full build (Python source + installable wheel + README + examples) is a one-time purchase — Individual $49, Team $149 (up to five named users):

https://payhip.com/Superkamoubot

Privacy note

Transcript content is escaped in the report, so nothing from an export can execute. The report still contains your session content — review it before sharing with anyone.

---

Disclosure: we publish Agent Run Inspector. The screenshots in this post come from the synthetic sample pack, not from a customer's session.