From ce0def14e7f3fd56c4e8646d17e5c206c589db8c Mon Sep 17 00:00:00 2001 From: Jason Woltje Date: Tue, 13 Oct 2026 07:48:24 -0500 Subject: [PATCH] agents, decisions: fresh Filbert Codex thread f712d8af for non-sandbox reviews, Dewey reviews sandbox rows (decision 130) Co-Authored-By: Claude Opus 5.5 --- agents/README.md | 2 +- docs/plans/2026-09-26_lead-decisions.md | 29 +++++++++++++++++++++++++ 2 files changed, 30 insertions(+), 1 deletion(-) diff --git a/agents/README.md b/agents/README.md index 0fc6090a0..bc9a8ae7b 100644 --- a/agents/README.md +++ b/agents/README.md @@ -14,7 +14,7 @@ not fleet seats. Keep changes in `/mnt/storage/src/mosaic-stack`; do not modify | Darkwing | Hands-on engineering; collaborating seat under Sage | Pi, configured Mosaic model | `agents/darkwing/launch.sh` | | Dewey | Frontend design, UX, accessibility, and UI implementation | Pi, configured Mosaic model | `agents/dewey/launch.sh` | | Rocko | General development, investigation, testing, and review | T3 session (Codex, gpt-6.1-sol, thread d99326cc, lead decision 122; Claude thread b84bb264 retired 2026-10-12); launcher Claude Code, Sonnet model | `agents/rocko/launch.sh` | -| Filbert | General development, investigation, testing, and review | T3 session (Codex, gpt-6.1-sol, thread 50342f6a, lead decision 122; Claude thread 9cb9731e retired 2026-10-12); launcher Pi, `openai-codex/gpt-6-astra:low`, not to be used under R26 until it changes (lead decision 69) | `agents/filbert/launch.sh` | +| Filbert | General development, investigation, testing, and review | T3 session (Codex, gpt-6.1-sol, thread f712d8af, non-sandbox reviews only, lead decision 130; Codex thread 50342f6a retired 2026-10-13, Claude thread 9cb9731e retired 2026-10-12); launcher Pi, `openai-codex/gpt-6-astra:low`, not to be used under R26 until it changes (lead decision 69) | `agents/filbert/launch.sh` | | Researcher | Source-grounded technical research and evidence | T3 session (Claude, Sonnet 5.5, thread 9bac0794, rows 97 and 98, lead decision 124); Haiku rows (71, 80, 84, 88, later 96) run as subagents Sage dispatches with model haiku, effort high and a worktree (lead decision 129); Haiku thread c98aaf4a retired 2026-10-13 unused; launcher Pi, configured Mosaic model | `agents/researcher/launch.sh` | | Sage | Project lead: coordination, review, integration; earlier DYOR strategy records retained | T3 session (Claude Code); Pi launcher `zai/glm-5.3:high` retained | `agents/sage/launch.sh` | diff --git a/docs/plans/2026-09-26_lead-decisions.md b/docs/plans/2026-09-26_lead-decisions.md index 4d47fd46d..c351cdb19 100644 --- a/docs/plans/2026-09-26_lead-decisions.md +++ b/docs/plans/2026-09-26_lead-decisions.md @@ -2907,3 +2907,32 @@ which stay with him. Each item names who decided it and what happened. - The Haiku T3 thread c98aaf4a ran at T3's default medium effort. It got only a holding prompt and is retired unused. Sonnet Researcher 9bac0794 keeps rows 97 and 98. + +130. **Sandbox rows leave the Codex reviewer; a fresh Filbert thread takes + the rest (2026-10-13, Mos's ruling on Sage's request).** Filbert's Codex + thread 50342f6a stopped twice on OpenAI's cyber-risk filter: at 00:12Z, + mid-way through the row 91 round 2 review, right after reporting "The + probe reproduced a host command escape. A nested submodule's clean", + and at 12:35Z, before it could write that finding up. The thread's own + history holds the flagged text, so every retry trips it again. + - Filbert moves to a fresh Codex thread on gpt-6.1-sol, f712d8af, + launched under decision 42. 50342f6a is retired. The new thread + reviews rows outside the seat sandbox, launcher and escape scope: + 97, 98, 71, 84 and 85 now, and later rows of that kind. Row 97 + round 2 goes to it first. Nobody pastes the old thread's flagged + text or any probe output into it. + - Rows 91 and 93, and any later row with sandbox or escape scope, + get their second review from Dewey on Claude. No partial GPT check + on those rows. This changes decision 122's review routing for them. + Row 91's reviewer is now Dewey (queue revs 881 to 883). + - Nobody applies for OpenAI's "Daybreak" access, and Mos isn't + raising it with Jason. + - Filbert's finding counts as a real candidate defect in row 91 until + Darkwing disproves it. Row 91 doesn't land, and the bus restart + doesn't include it, until round 3 closes it with a test or Darkwing + shows it doesn't reproduce. If row 91 is still open at the PM + standup, it counts as stuck under decision 118, and Mos decides + then whether the redeploy ships without rows 91 and 93. + - Row 93 is owned by Filbert as builder, which the ruling didn't + cover. It stays parked behind row 91, and Sage asks Mos who builds + it.