Anti-pattern log: a historical SDK inventory is not the pin oracle for a live SDK (Codex runtime lane handoff)

[seathatflowsinourveins]

Scope

• What this PR changes, in one or two sentences: adds one dated row to the anti-pattern log in docs/harness-defaults.md, handed to the coordinator by the Codex runtime lane: a historical SDK supply-chain inventory must not become the pin oracle for a live SDK after a supported upgrade.
• Base commit: origin/main after Rule text in every layer: the six standing clauses on AGENTS.md, the portable Claude template and the Codex AGENTS template, anti-pattern rows and the folded finalization record (unit F1) #557 (the rule-text unit, which owns the log's other 2026-09-30 rows).
• Lane: lane:foundation
• Owned paths touched: docs/harness-defaults.md; manifests/evidence.json (last commit only, re-registration of the changed hash-listed file).
• grants: none

SOTA sources

• The Codex runtime lane's handoff (2026-09-30 23:4xZ, corrected 23:56Z): GitHub Actions run 36771924863 / job 110080191631 (main control green, draft red), the repaired tests/test_adoption_contract.py and evidence/artifacts/runtime-sdk-20260930/ci-repair-receipt.json at PR Publish Codex SDK 0.159.2 route qualification and scoped usage accounting #560 head 74902ff1; the historical SDK 0.154 inventory retains 34 non-SDK dependency versions; the live 0.159.2 SDK/CLI truth is the independent canonical pin and qualification guard (openai/codex rust-v0.159.2).
• AGENTS.md: "record each proven mistake in its anti-pattern log"; docs/harness-defaults.md#upstream-verification-and-compounding-learning.

Evidence-class table

Claim	Evidence class	Command / receipt
The log's table stays well-formed and the docs stay consistent	local_integration	python3 -m unittest tests.test_adoption_docs_consistency tests.test_install_claude_profile (104 OK, skipped=4)
Hash registration	local_integration	python3 scripts/validate.py passed

No unchanged upstream test suite was run; the row records the Codex lane's own CI observation, cited above.

Local commands run

$ python3 -m unittest tests.test_adoption_docs_consistency tests.test_install_claude_profile
exit 0: Ran 104 tests; OK (skipped=4)
$ python3 scripts/validate.py
exit 0: "status": "passed"

Decision record

None; the row is the record.

Host evidence

Not applicable.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets printed, logged or committed.
• No paid hosting, subscription or billing surface.
• Peer-owned untracked files and worktrees preserved.

🤖 Generated with Claude Code

[seathatflowsinourveins]

Cross-family review of the whole PR (GPT-6.1 Sol at max on the native Codex login, read-only codex exec, web search off, 2026-10-01 ~01:05Z): verdict approve, no findings ("CI logs and PR #560 at 74902ff support the record; checksum, publication validation, and 12 adoption tests pass"). This PR touches the hot file (re-registration of docs/harness-defaults.md), so it merges after the trading lane's #559 and the remaining train PRs to spare them another hot-file re-run.