Close out the 11 reviewed transcripts; add the 2026-08-31 team conversation

[jon-olson-tylertech]

Step (g) of the review process for the 11 transcripts sitting in reviewed.

What this does

	Count	Why it closes
action_status: none-needed	9	answer_verdict: good, fix_target: none — nothing was ever owed
action_status: applied	2	knowledge-file changes that already shipped (07a6ae94, 9c230f8d)

pushed is a claim about Foundry, not about the repo, so it is made against a measurement rather than an assumption: python3 scripts/check_foundry_drift.py reports all five collections and the team router byte-matching main as of 2026-09-02. Each transcript's notes field records that, so the basis for the claim is readable later.

Nothing was reclassified and no verdict was altered — the only frontmatter change is review_status: reviewed -> pushed plus the appended note. reviewer is untouched on all 11.

Also in here

• transcripts/team/2026-08-31--12233958.md — pulled by fetch_transcripts.py on 2026-09-01 and left uncommitted. Still pending; it is one of the 7 awaiting a human.
• transcripts/INDEX.md regenerated.

Why it matters

reviewed was holding 11 items nobody owed anything on, which buries the 7 that do need a human. In-scope close-out goes 47% -> 79%.

Checks run locally

validate_reviews.py (11 first reviews, 0 problems) · review_status.py --check · check_admin_paths.py · gen_codeowners.py --check · check_folder_ownership.py — all pass.

Not in scope, but noted while looking

All three new bp-general transcripts (2026-08-28) show "answered without searching" — zero tool calls on all three. That is no-search, an agent-prompt problem rather than a knowledge-file one, and it is the corpus's first three real conversations. Flagged for whoever reviews them; not touched here.

🤖 Generated with Claude Code

---

⚠ Two problems found while opening this PR — reported, deliberately NOT fixed

Both live in admin-only paths, and this PR's author is a contributor (write, role reviewer), so they are not his to change.

1. Knowledge-Shared/ is admin-only everywhere except the file that enforces it.

CLAUDE.md rule 6 names Knowledge-Shared/ as admin-only and explains why — its files upload to all five collections, so a change there alters what every agent says. .github/CODEOWNERS:27 enforces it (@onetyler-tcp-pm-admins, no contributor). But it is absent from .github/admin-only-paths.txt, which that file's own header calls the "SINGLE SOURCE for the boundary".

Consequences:

• .github/workflows/validate.yml reads admin-only-paths.txt, so the CI tripwire would not fail a non-admin PR touching Knowledge-Shared/.
• scripts/start_review_session.sh would not warn before the commit.
• scripts/check_admin_paths.py reports "CODEOWNERS and admin-only-paths.txt agree on all 16 admin-only path(s)" — it passes, so the check that exists to catch exactly this disagreement does not catch it.

CODEOWNERS review is currently the only control. That is the one control CLAUDE.md says cannot be relied on alone.

2. allow_auto_merge is false on this repository.

CLAUDE.md states every request this flow opens is put into auto-merge at creation, and that a failure there is a permissions problem an admin turns on. It is not permissions — it is the repository setting:

$ gh pr merge 96 --auto --squash
GraphQL: Auto merge is not allowed for this repository (enablePullRequestAutoMerge)
$ gh api repos/tyler-technologies/onetyler-foundry-team-agent-kb -q .allow_auto_merge
false

So gh pr merge --auto cannot succeed for anyone until "Allow auto-merge" is enabled in Settings. Until then the documented Blueprint pairing — knowledge file and Blueprint page merging together without someone returning at the right moment — does not hold, which is the exact failure auto-merge was introduced to prevent.

[vijay-tylertech]

INDEX.md regenerated in a6d359f — the header line was the only casualty of the merge resolution, and validate is green. Transcript verdicts and the drift-verification notes reviewed and unchanged.

[vijay-tylertech]

Re-approving after the rebase onto main. Two conflicts resolved: INDEX.md regenerated (not hand-picked), and transcripts/team/2026-08-31--12233958.md resolved to the complete copy — main's was missing Exchange 3 — with the text typed into review:2 preserved. validate green, validate_reviews reports 11 first reviews and 0 problems.