docs(batch-review): reviewer-owned batch QA with sub-agents - #45
Merged
Conversation
Add references/batch-review.md: how one reviewer takes a batch of QA tickets through a single Round-1 pass by fanning the evidence-gathering out to sub-agents while every claim, verdict, comment and transition stays the reviewer's own. Ten-step sequence (claim all first, gather all into files, group per agent by domain, written briefing, agents draft / reviewer posts, settle every (?) yourself, "idle" is not a report, main comment before TL;DR chained with &&, stop every agent and remove what it left, log time once per batch), a sizing note (fan-out from roughly six tickets), the generalized briefing template, and a table of what stays with the reviewer. Cross-link from lifecycle.md Stage 0 in one sentence, add the batch-review.md index entry to SKILL.md (499 words, under the 500 cap; one filler word trimmed from the Output section to make room), and add eval E10 to evals/qa-discipline.md in the existing stub format. Observed in the 2026-08-21 /retro of a Round-1 IT QA session: 16 maintenance tickets, 5 sub-agents grouped by component (2-4 tickets each), all 16 passed. Every (x)/(!) in the agent drafts was re-measured by the reviewer before posting (a CHANGELOG-regression claim via `git show <tag>:CHANGELOG.md | grep -c`, stale directory PTRs re-dug, a "device alive" claim re-pinged, a CI schedule's variables re-read via the API); a `;`-chained post once published a TL;DR after the main comment had been rejected; one agent went idle without a report while its deliverables were already on disk; agent processes and clones had to be swept explicitly at the end. Learning-Id: retro-20260821-qa-batch-fanout Assisted-by: claude-code:claude-fable-5 Agent-Session: https://claude.ai/code/session_0124WZK5UoAM9oYQoSYF1Y8z Agent-Host: 32116e Signed-off-by: Sebastian Mendel <github@sebastianmendel.de>
|
Member
Author
|
Self-review: 7a168f7 The review this pull request demands is unsatisfiable (Copilot quota wall or repeated bot failures on this head). Per the documented fallback, the diff on this head was reviewed by the PR author; this comment is the on-the-record attestation the merge gate reads back. It stops matching on the next push. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds
references/batch-review.md: how one reviewer takes a batch of QA tickets through a single Round-1 pass by fanning the evidence-gathering out to sub-agents while every claim, verdict, comment and transition stays the reviewer's own — a ten-step sequence (claim all tickets first under the reviewer's account; gather every Stage-0 bundle into files before fanning out; group 2–4 tickets per agent by domain/component; a written briefing every agent reads first; agents draft, the reviewer re-measures every load-bearing(x)/(!)and posts; every(?)in a draft is settled by the reviewer or downgraded to(!)with the reason, never posted; "idle" without a report means read the deliverable files; main comment first, TL;DR only on success, chained with&&; stop every agent and remove clones/scratch, ending with zero agent processes; log the review time once per batch), a sizing note (fan-out from roughly six tickets, below that inline), the generalized briefing template (vendor-neutral placeholders:<ticket-bundle-dir>,<jira-cli>,<forge-cli>,<host>, …), and a table of what stays with the reviewer. Cross-linked fromlifecycle.mdStage 0 in one sentence; one index entry added toSKILL.md(499 words, under the 500 cap — one filler word trimmed from the Output section to make room); eval E10 added toevals/qa-discipline.mdin the existing stub format.Why
Observed in the 2026-08-21
/retroof a Round-1 IT QA session: 16 maintenance tickets reviewed with 5 sub-agents grouped by component, all 16 passed. What made it sound — and what nothing in the skill yet said — was that the agents only gathered evidence and drafted, while the reviewer claimed, re-measured every load-bearing claim before posting (a CHANGELOG-regression claim viagit show <tag>:CHANGELOG.md | grep -c, stale directory PTRs re-dug, a "device alive" claim re-pinged, a CI schedule's variables re-read via the API), settled the drafts'(?)items, posted and transitioned. Two frictions from the same session are encoded as rules: a;-chained post once published a TL;DR after the main comment had been rejected by the markup validator, and one agent went idle without a report while its review files were already on disk; agent processes and clones also had to be swept explicitly at the end.How to verify
bash <skill-repo-skill>/skills/skill-repo/scripts/validate-skill.sh .→ 0 errors,SKILL.md is 499 words (under 500 limit);check-version-parity.sh→ versions match at 0.7.0 (no bump, feature PR);pre-commit run --files <changed>green (markdownlint, whitespace, EOF). Readreferences/batch-review.mdagainstlifecycle.md— no stage is added or reordered, only sequenced across many tickets.Came from /retro: yes
Learning-Id: retro-20260821-qa-batch-fanout
Assisted by claude-code:claude-fable-5 — Session