You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit e520a67
Browse filesBrowse the repository at this point in the historyBrowse files
Let harness see every agent event and stop retrying what fails the same way
The agent wrapper only forwarded the session event, so RunSession never saw a lost conversation and never restarted it fresh, and StreamCmd never saw the error events it classifies as account errors. The retry now skips a model account refused for good and a spent budget. Codex gets its effort through Job.Effort, catalogued CWE ids such as cwe-787 are spelled canonically instead of dropped, GitHub reads retry 429, 502, 503 and 504, a test keeps the agent image pins in step, and CI vets the evals.
Copy file name to clipboardExpand all lines: docs/configuration.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -56,7 +56,7 @@ Each report names the source commit and the input revision it covers.
56
56
| Request | 1 MiB of body. API responses ask for at most 16,000 output tokens, and the ChatGPT backend applies its own output limits. |
57
57
| Reports | One per known reporter and per execution |
58
58
59
-
tamis starts a report only while the execution still holds a whole advisory budget and thirty minutes of its time. Otherwise the report waits for the next execution. SDK retries are off, but tamis retries an agent run that fails without a valid report once, after a minute. When a budget runs out, the report stays retriable and tamis does not mark it complete.
59
+
tamis starts a report only while the execution still holds a whole advisory budget and thirty minutes of its time. Otherwise the report waits for the next execution. SDK retries are off, but tamis retries an agent run that fails without a valid report once, after a minute, unless the model account was refused for good or the advisory budget is spent. A GitHub read answered with 429, 502, 503 or 504 is tried up to three times. When a budget runs out, the report stays retriable and tamis does not mark it complete.
60
60
61
61
These bounds cap one execution, not the spending across scheduled runs. Provider pricing, tokenisation and ChatGPT credit policies differ, so request and byte limits give no monetary guarantee. On the provider side:
0 commit comments