Repository navigation
Archive every model call as encrypted, deduplicated daily traces with verified off-site upload - #12
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Every prompt, reasoning trace and tool call that passes through the hub was thrown away. The only way to keep them was
request-log, which writes debug dumps at about 1.2 GB per hour because agent clients resend the whole conversation on every turn. This PR adds a long-term trace archive. It keeps a durable record of every model call for evals, fine-tuning, and replaying old prompts against future models.What gets recorded (opt-in;
trace-archive.enableddefaults tofalse):redacted_thinking, tool calls, and Responses reasoning items withencrypted_contentstored verbatim so the conversation can be replayed.How it's stored:
authorizationandauthorization_tokenfields and credential headers are masked before hashing.<auth-dir>-traces, outside the credential tree.Tools:
t3code-proxy traces export|verify|statscommands./v8/management/trace-archive/{settings,status,seal-now}endpoints;/v0is unchanged.docs/trace-archive.mdspecifies the format precisely enough to decode with onlyage,zstdandjq.Proof:
make verifypasses, apart from two TUI tests that also fail onmainunder WSL; CI is the arbiter.tracearchive,apiand the Codex duplex executor./v1/messagescall was recorded and sealed.traces verifyandtraces exportreproduced the thinking block, its signature, the tool call{"city":"Seoul"}and tokens 42/37.ageCLI plus Python zstd decoded the segment, and the footer SHA-256 matched.Reviewed with sol-loop: 3 rounds with GPT-6.2 Sol (high). It found 14 issues; 13 are fixed, each with a regression test. Sol did not approve the final state because two narrow Codex duplex cases remain, listed below as follow-ups.
Known follow-ups (rare, and client traffic is unaffected):
Created with GPT-6.2 Astra (implementation, Codex) and Claude Opus 5.5 (coordination, review, verification) in T3 Code.
🤖 Generated with Claude Code