docs: add agent trace lessons

[mldangelo-oai]

Summary

• synthesize recurring lessons from ModelAudit Codex traces and current swarm evidence
• tighten exact-head PR closeout, focused CI diagnosis, and valid no-change guidance
• document bounded real-model QA, artifact inventory, fail-closed outcome checks, and standalone picklescan validation

Method

Five independent evidence lanes plus a skeptical deduplication pass were used. Instructions required corroboration from multiple traces or evidence strata; volatile campaign details and orchestration mechanics were excluded.

Validation

• documentation checks and formatting passed
• Ruff and mypy passed
• broad pytest exposed four unrelated xdist-only picklescan failures; each passes serially
• git diff --check passed against current main

[mldangelo-oai]

@codex review

Please review exact head 7e4ce87d for durability, duplication, contradictions, and transient/orchestration-specific guidance.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[github-actions]

Workflow run and artifacts

Performance Benchmarks

Compared 12 shared benchmarks with a regression threshold of 15%.
Status: 0 regressions, 0 improved, 12 stable, 0 new, 0 missing.
Aggregate shared-benchmark median: 1.435s -> 1.446s (+0.8%).

Workload	Benchmark	Target	Size	Files	Baseline	Current	Change	Status
warm-cache-rescan	tests/benchmarks/test_scan_benchmarks.py::test_scan_warm_cached_repository_rescan	release-candidate	547.3 KiB	32	95.91ms	99.71ms	+4.0%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_base64]	nested_base64	98 B	1	504.9us	519.0us	+2.8%	stable
direct-malicious-upload	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_direct_malicious_upload	malicious_reduce	52 B	1	452.6us	441.3us	-2.5%	stable
clean-training-checkpoint	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_clean_training_checkpoint	safe_large	278.2 KiB	1	111.98ms	113.43ms	+1.3%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_hex]	nested_hex	130 B	1	547.5us	554.5us	+1.3%	stable
padded-multi-stream-upload	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_padded_multi_stream_upload	multi_stream_padded	4.1 KiB	1	570.1us	564.3us	-1.0%	stable
nested-payload-review	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_nested_payload_review[nested_raw]	nested_raw	78 B	1	501.6us	505.3us	+0.7%	stable
mixed-model-repository	tests/benchmarks/test_scan_benchmarks.py::test_scan_release_candidate_repository	release-candidate	547.3 KiB	32	480.26ms	483.14ms	+0.6%	stable
suspicious-pickle-intake	tests/benchmarks/test_scan_benchmarks.py::test_scan_suspicious_pickle_intake	suspicious-intake	183.8 KiB	4	143.49ms	144.34ms	+0.6%	stable
duplicate-heavy-registry	tests/benchmarks/test_scan_benchmarks.py::test_scan_duplicate_registry_snapshot	registry-snapshot	915.2 KiB	13	409.22ms	411.42ms	+0.5%	stable
chunked-upload-stream	tests/benchmarks/test_picklescan_benchmarks.py::test_picklescan_chunked_upload_stream	chunked_stream	278.2 KiB	1	116.70ms	116.43ms	-0.2%	stable
single-checkpoint-preflight	tests/benchmarks/test_scan_benchmarks.py::test_scan_single_checkpoint_before_load	single_checkpoint.pkl	183.0 KiB	1	75.19ms	75.26ms	+0.1%	stable

[mldangelo-oai]

CI recovery note for exact head 7e4ce87d57856d770f3930584056408bfea67f28:

• Initial Docker run 27313027781 built the full image and passed all container smoke tests. Trivy alone failed while copying a layer into /tmp/trivy-5672/... with no space left on device.
• The PR changes only AGENTS.md and packages/modelaudit-picklescan/AGENTS.md; it does not change Docker, workflow, dependency, or lock files. The same base SHA (2f782ba1) also passed Docker Image CI on main in run 27312571374.
• I reran failed jobs only. Attempt 2 passed the full-image job and the Docker CI Success gate on the unchanged head.

Conclusion: GitHub-hosted runner disk-exhaustion flake; no code or workflow change was warranted.

[mldangelo-oai]

@codex review

Please review exact head 5e376382c27ded6885a09ea7c82197fd66fd8782 after the additive merge from current origin/main. Focus on durability, duplication, contradictions, transient/orchestration-specific guidance, and merge-result regressions.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".