chore(seidroid): take uci v0.0.22

[bdchatham]

What this is

Repoints this repository's seidroid review caller from a raw commit sha to the uci tag v0.0.22.

seidroid-review.yml landed on uci's main in sei-protocol/uci#108 and shipped as v0.0.22, so for the first time there is a tag to name. Until now every caller pinned a sha on a branch that had to be kept alive by hand.

What changes here

The three uses: lines, and nothing else. No inputs change.

This caller omits driver-version, so it inherits the workflow default, which v0.0.22 sets to sei-agent-driver/v0.21.0. Two things follow from that:

• The review summary comes first, entries are one or two sentences, and nits are folded into a collapsed block (PLT-1268).
• The verdict rides in the review that carries the inline comments, instead of a separate comment beside it. A standalone comment is now only the fallback for a run where that review did not land.

This repository was further behind than the others

It pinned 92f74c9, the v0.18.0 driver-floor commit, where the other four pinned 81e975c. So it also picks up everything between those two: the recorded review decision and the base branch's Accepted list, run-deadline-seconds, and the PLT-1268 change above.

The header comment changes too

It said the uses: lines pin a raw sha because "this callee is on no uci tag and on no uci default branch... so there is no tag to name." That is no longer true, so the pin stops being one of this file's documented divergences — four become three, and the paragraph explaining the sha is replaced by one sentence saying the lines pin a tag, as this repository's four other uci callers already do.

Known, non-blocking

Four items are tracked for the next uci release and listed in the release notes — the most visible is that a head which moves mid-review still earns a 👍 on the trigger comment, even though the run correctly declines to approve or to clear a block. None gates a merge.

mode in the concurrency group lets a review and a close run together for one pull request. The trade-off is deliberate and stated in the callee; PLT-1274 tracks the driver-side question.

🤖 Generated with Claude Code

[cursor]

PR Summary

Low Risk
CI wiring only—reusable workflow ref and comments—with no application or secret-handling changes in this diff.

Overview
Repoints the seidroid review reusable workflow from a pinned commit (92f74c9…) to sei-protocol/uci@v0.0.22 on all three jobs (seidroid-review, seidroid-review-close, seidroid-review-reclaim). Job if conditions, secrets, and with: inputs are unchanged.

The file header is updated to match: uses: now documents tag pins like the other uci callers, so the “four deliberate divergences” note drops the sha-pin case (three remain: lowercase name, empty top-level permissions, no concurrency).

Because this repo does not set driver-version, moving to v0.0.22 also pulls in that release’s default driver and review UX (e.g. summary-first output, verdict on the inline review with a comment-only fallback), plus anything shipped between the old sha and the fleet’s newer pin.

^{Reviewed by Cursor Bugbot for commit 72cd1cf. Bugbot is set up for automated code reviews on this repo. Configure here.}

[seidroid]

review found nothing blocking.

[seidroid]

Repoints all three seidroid review job uses: lines from the raw uci sha 92f74c9 to the tag v0.0.22 and rewrites the header paragraph that justified the sha. Verified against the callee at v0.0.22: the tag exists and carries seidroid-review.yml, no input this caller passes (mode, approve-on-success, allowed-team, guidelines-file) or secret was removed or retyped, the job-level scopes the callee's two jobs declare are exactly the four this caller grants, and the only default shifts reach inputs this caller deliberately omits (driver-version v0.18.0→v0.21.0, timeout-minutes 45→60, new run-deadline-seconds 1800) — the sole reading, codex, contributed nothing because its session failed, so nothing was carried or dropped from it.

Non-blocking

• The header no longer records that this file moves in lockstep with the other repositories wiring this reviewer, which was the stated reason behind the retained "keep the three occurrences below identical". The PR body says the fleet is not currently in lockstep, so nothing is wrong today, but the next bump loses the note that it is a fleet-wide, path-driven change.

1 nit, not posted on the code

• .github/workflows/seidroid.yml:102 — A tag is a movable ref, so all three jobs now resolve whatever v0.0.22 points at rather than a fixed tree — and this caller is the one that hands the callee the App private key, the machine-client secret and a shell allowlisted to Bash,Read, where uci-lint and the other three tag-pinned callers risk far less. Same-org and matching this repository's convention makes it a defensible trade; noting it only because the blast radius here is not the siblings'.

_{seidroid review · decision approve · session 7f1ada43bc0b487a975f2ce0f28b8e42 · turn resp_claude_9331c4b3aea916b7bf7b0ef7fffba008 · item eafae8b80e605e16ad2f4a84bd9497d0}

Findings: 0 blocking | 1 non-blocking | 0 posted inline