Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 33 additions & 0 deletions quest/m0/fork-review-record.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
# [S] quest review records on a contributor's fork PR

## Goal

`quest review` in a maintainer's checkout of a contributor's fork PR finds
that PR, reads its `## Models` size, and writes its `## Review` record, so
nobody writes one by hand. Contributors reviewing in their own fork's clone
keep working as today.

## Plan

Decided 2026-11-17 (maintainer, `/quest-plan` after `/quest-run`): ✅ m0
(rejected: m1), since every fork PR's local review now needs a hand-written
record that `/quest-merge` then trusts; ✅ fix the lookup (rejected: a
`--pr <url>` flag; dropping it).

- Cause: `openPull` in `src/cli/review-record.ts` matches only an open PR
whose head repository is the one the base's remote pushes to. In a
maintainer's checkout that is the upstream, not the contributor's fork, so
it prints `record skipped: no open PR from ...`, and `pullModels` returns
null, so the round runs at the default size. Both iterates on
[#761](https://github.com/moq-dev/quest/pull/761) hit this: they reviewed
at size M instead of the body's S and wrote the record by hand.
- Those iterates worked in a detached checkout with `--branch` and pushed to
the fork by URL, so the checkout has no push remote naming the fork, and
the reviewed head is not pushed yet when the round ends.
- Recommended: match the open PR into the base whose head branch is the
reviewed branch, from any head repository; prefer the push repository's
when several match, and skip naming every candidate when the choice is
still ambiguous.
- Test: a stub forge with an open PR from `contributor/quest:<branch>`
into the upstream gets its record rewritten and its size read from a
maintainer's checkout whose remote is the upstream.
30 changes: 30 additions & 0 deletions quest/m1/codex-146-messages.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
# [XS] quest messages reads Codex 0.146 sessions

## Goal

`quest messages` reads interactive Codex 0.146 sessions, so
`quest messages --since 2026-10-01` succeeds on a host that ran 0.146 rather
than failing with "user turn without Codex's provenance". Turns of an
unknown shape from other versions still fail.

## Plan

Decided 2026-11-17 (maintainer, `/quest-plan` after `/quest-run`): ✅ read
0.146's turns (rejected: skipping any rollout quest cannot read with a
warning, which would let a future format change drop a host's data silently
and reverses [#719](https://github.com/moq-dev/quest/pull/719)'s choice to
fail on unknown shapes; both; dropping it); ✅ m1 (rejected: m0), since the
gap is two days on one host and does not recur.

- Codex 0.146 writes no `content_item_metadata` or `content_item_kinds` on
user turns, only `turn_id`. On one of the maintainer's hosts, 6
interactive sessions (2026-10-15 and 16) hold 33 such turns in the window,
and one aborts the whole run; that kept the host out of the transcript audit
behind [#764](https://github.com/moq-dev/quest/pull/764). Sub-agent
sessions of that version are already skipped by their object `source`.
- In those sessions, injected parts open with `<environment_context>` or
`# AGENTS.md instructions for`, and typed text is a plain part.
- Recommended: for a session whose `session_meta.cli_version` predates
provenance, treat a part opening with a known 0.146 wrapper as
harness or `agents_md` and the rest as typed, and keep throwing for a
turn without metadata from a version that writes it.
Loading