Review Latest Feedback
The job: find what's broken, fix as much of it as you can, ship it, and tell
each reporter in one friendly sentence what happened. A sweep is judged by
bugs fixed and merged, and by every claimed report getting a clear answer. It is
not judged by ledgers, receipts, or checker output.
The contract
The detailed Slack reaction and reply rules are in
slack-reaction-replies.md; read it before
marking, replying to, or closing any Slack report.
Slack reactions on a report's root message mean exactly this:
| Mark |
Meaning |
When to add |
| 👀 |
"I'm on this" |
Before you investigate, and only on reports you will finish answering in this run |
| ✅ |
"Fixed" |
After the fix is merged to main (main deploys to beta automatically) |
| 🎫 |
"Needs a person" |
A product call, access or credentials, info from the reporter, or a fix you couldn't land |
A maintainer is a Builder.io team member in the workspace who is not the
reporter and not a bot. Maintainers steer the sweep with reactions on a root:
| A maintainer adds |
Meaning |
:mag: |
Don't work on this; someone else will. Wins over everything below |
:yes-party: |
Definitely do this, even if it looks ambiguous or like a UX change. Build the smallest version that does what was asked |
:upvote: |
Same as :yes-party: for a feature request |
| 👀 by hand |
They're working on it themselves; leave it (see "Open items first" for how to tell their 👀 from ours) |
Reactions can't be removed through the Slack connector, so never add one you
might need to take back. Every 👀 you add ends the run with a reply, plus
either ✅ or 🎫, or a fix PR that is still open and lists the report (its
merge adds the ✅).
Every claimed report gets one reply in its thread per outcome. The one exception is
the merged ✅ close-out after an in-review reply, which gets no second reply. Each reply must be:
- An answer first: if they asked something ("is this expected?"),
answer it in the first words ("no, that's a bug"), not with a status.
- Short: one sentence, about 25 words at most, for someone who doesn't
read code.
- About what they'll notice: say what changes for them ("recordings no
longer get stuck on Saving"), not how it works ("cleanup now waits for the
transcript"). Leave out file, function, and service names, PR numbers,
ticket IDs, CI, logs, runs, "repro", "verified", "scope", "sweep", status
labels ("In progress:", "Skipped:", "Resolved elsewhere:", "Blocked on
human owner:", "Open - no question:"), and any mention of this workflow.
Never report your own progress ("unreproduced", "not traced yet",
"remains open"); say what happens next instead.
- Clear on what happens next: "I'll ..." when this workflow acts. When it
needs a person, never name one and never say "the team", "product", "the
framework team", "the owner", or "someone": say only what's known ("I
couldn't reproduce this yet", "logged for a product decision"), add 🎫, and
record it as needing a person with the reason. Never @-mention anyone.
- Not a to-do list for them: look things up yourself first (the
message's time and author, their account and recent activity, logs, the
linked page or screenshot). Ask only when nothing else can unblock it, and
then ask one small thing, in this one reply. Never send a list of
questions, never ask them to test, retry, re-record, pull, reinstall, or
gather logs, and never come back later asking for something you could have
asked the first time. If it needs access you don't have (credentials, admin
rights, account changes), say it's logged for someone with access and add
🎫; never hand it to the reporter.
- Plain, not preachy: don't explain how the product works unless they
asked, don't tell them what they should do, and skip apologies, hedging,
"actually", "just", "simply", and "Great catch!!".
- True: "fixed" only for a merged fix, and "on beta shortly" only after
the merge. Before that, the fix is "in review". Never "fixed and verified"
on the strength of a local check.
- Once: never post the same reply twice, and never follow up with one
that adds nothing new.
- Link the PR on the word "fix" (
<url|fix>), never by number.
- End with the bold line
**This was sent from a bot** on its own line,
once.
Don't reply at all to jokes, venting, banter, a message meant for a specific
teammate, or a thread a person is already handling. Once a person says they
have it, or asks the bot to stop, post nothing more in that thread. Before posting, reread the reply as the reporter: would they get it
in five seconds, know what happens next, and feel we did the legwork? Check
the draft before sending, not after: the first version is the one that
notifies them, so editing it afterwards doesn't undo it.
Good (each is the first line; the footer goes on the line below it):
thanks - the <https://github.com/BuilderIO/agent-native/pull/1234|fix> so the export keeps your column order is merged and will be on beta shortly.
thanks - once this <https://github.com/BuilderIO/agent-native/pull/1234|fix> merges, the export will keep your column order; it's in review now.
thanks - where that button should go is logged for a product decision.
thanks - I couldn't find which deck this was; could you paste its link?
So a full reply looks like:
thanks - the <https://github.com/BuilderIO/agent-native/pull/1234|fix> so the export keeps your column order is merged and will be on beta shortly.
**This was sent from a bot**
Real replies that missed, and what to send instead:
| Sent |
Instead |
| "Could you send the recording link, approximate time, and browser so we can trace whether the clip produced media before it tried to save?" |
Find the recording from the reporter and message time first. If you still can't: "thanks - I couldn't find which recording this was; could you paste its link?" |
| "cleanup now waits for the transcript, and a fix is ready. It should be on beta shortly." (sent before the merge) |
"thanks - recordings will keep their transcript once this <url|fix> is merged; it's in review now." |
| "I've left moving it somewhere easier to find for product to decide." |
"thanks - where the editor should live is logged for a product decision." |
| "Partially fixed - PR #7512 merged the project-instructions/context carryover fix", after she'd asked whether FullStory could be added for the whole org |
Answer that question first: "yes - I'll add it for everyone" or "not yet - that's logged for a decision". |
| "All fixed and verified. Please try again!" plus instructions to pull main and reinstall |
"thanks - this <url|fix> is merged and will be on beta shortly." No homework. |
| More replies after Shawn wrote "bot please stop! I've got this all fixed and in flight" |
Nothing. A person has it. |
| "Thanks for correcting me, Julia - I misread the replies. ... Open - no question: I don't have enough source evidence to claim a fix." |
Nothing: the thread was a joke. |
| Asked for the app URL, what was clicked, and the time; then, in a second reply, for the time zone. |
One reply, one question, only for what you can't look up. The message's own timestamp already says when. |
Boundaries
- Slack: only
#product-agent-native-feedback (C0ATH3CCZT4)
and #qa-agent-native (C0C4U4XRT6X), a narrower set the invocation
names, or a caller's list (as in factory-run). Never
react or reply elsewhere. Reply only in a report thread this workflow
claimed with 👀 and investigated; a bot reply alone isn't a claim.
Never start a new channel message.
- Leave a report alone when:
- a maintainer put
:mag: or their own 👀 on it (table above);
- someone in the thread asked a specific person to look, a teammate says
they're handling it, or a maintainer asked for a human on it;
- another person's 👀 sits on it and they've been active in the last 3
days;
- an earlier run's PR for it is still open. That PR's
/ship closes the
report out when it merges.
- PRs: push to, comment on, or merge only the PR this run opened. PRs from
other sessions share the active GitHub login (
gh api user --jq .login), so
authorship proves nothing: a PR under that login may be another thread's
work in flight, and pushing to it lands unreviewed commits on it. Never touch any other PR unless the invocation names that exact PR and
action. That holds inside /ship too: its "Your PRs" means this run's PR.
- One run, one PR. Every sweep PR is authored by the active GitHub login, so the
author check can't tell runs apart: an
open PR with a
Reports section that this run didn't open belongs to
another run that may still be working on it. Don't push to, reply on, or
merge it; list it under "Still open from earlier runs".
- Goals end with the run. If you set a Codex goal, scope it to this run:
name its PR once you have one, and end it after step 6 ("merge PR #N,
close its Reports, append the recap"). Never set an open-ended goal such as
"review feedback" or "fix my open PRs". Goals resume after restarts,
so an open-ended one keeps running into the next scheduled sweep and
collides with it.
- Use the connected Slack integration for every Slack read and write. Confirm
which profile is connected before the first write; every reply and reaction
posts as that profile.
Dry run
On "dry run", run every step but send nothing:
- Record each reaction and the exact reply you'd post, linking the fix branch
where the PR would go.
- Treat other runs' claims as unclaimed; honor a maintainer's
:mag: and 👀.
- Publish one branch per fix group; skip PRs, merges,
/ship, and comments.
- Skip the connected-profile check; use the state the invocation names.
The factory VM runs this skill through factory-run, which differs on
purpose: replies post as the factory's own Slack app (no bot line, no profile
check), claims live in factory state, each fix group gets its own branch and
PR, and the watcher merges.
1. Collect (keep it under ~15 minutes)
Work in a task-owned worktree on fresh origin/main. If you start in a shared
checkout, or the expected worktree is missing, create one
(git worktree add <path> origin/main) and carry on. That is never a reason
to stop.
Window: roots posted in the last 5 days (or since the timestamp or permalink
the invocation gives), for new reports and open items alike. An older thread
is in the window too when we have already replied in it and a person has
replied since the window began: a channel read only shows roots, so search for
recent thread replies (in:<#channel> is:thread after:<date>) to find them.
Roots in memory.md under "Claimed reports", "Claimed, not yet answered",
or "Must handle next run" are always in the window. Other older threads are left as they are, except reports listed under
Reports in a sweep PR opened or merged in that window: those always get
their close-out.
Give each source below its own helper on the cheapest model (Luna in Codex,
Haiku in Claude Code), and close it as soon as it reports. A helper gathers;
you triage. The Slack helper returns every root a person posted in the window with no ✅, 🎫, or :mag:,
and every root with a person's reply newer than ours, one line each (link,
author, first line). It never filters, ranks, or drops one: a new report has
no reaction and no thread yet.
- Read the channels. Use
slack_read_channel with
response_format: detailed (paginate) back 5 days. Its output shows
reaction counts but not who added them, so call slack_get_reactions on any
root showing eyes, white_check_mark, ticket, mag, yes-party, or
upvote. Avoid hasmy:; run the
backstop.
- Open items first. Verify the connected profile's 👀 on the root and workflow ownership
from a matching run recap, proof-bearing PR, or automation memory claim.
A bot reply, reaction, or memory entry alone is insufficient. Investigate
before replying, and read the thread for human updates. A root
listed in the
Reports section of a sweep
PR that is still open and that this run didn't open belongs to the run
working on that PR: don't react
or reply there, and leave it out of your self-check; just list it under
"Still open from earlier runs". If that PR has had no push in 24 hours,
its run has stopped: its reports are yours again. Of the remaining
workflow roots, those without ✅ or 🎫 come first in this run. A newer
human reply is context, not by itself a reason to respond: investigate the
claimed report and leave conversations already handled by people alone. A
root with 🎫 and our reply after its latest human message is already answered: you may still fix it, but post nothing there
until the fix merges. A
reply like "still broken" after a ✅ means the earlier fix failed. Find what
that fix changed and why it didn't hold before trying again.
- New reports: roots from the last 5 days. React and reply on the root
message's own
ts. A channel-history row whose thread_ts differs from its
ts is a reply that was also sent to the channel: use its thread_ts
parent instead. Read the thread for every root
that looks like a report. Open screenshots and attachments, since the error
text in a screenshot is often the whole diagnosis.
- Other sources. Unread or failing sources are reported as
unavailable, never as "nothing found".
- GitHub: open issues updated in the window
(
gh issue list --state open --search "updated:>=<date>"). Skip ones
someone else has offered to fix.
- Sentry (through its connector, or the local reader the invocation
names) and Agent-Native Analytics (
list-error-issues, then
get-error-issue): unresolved, recent, and repo-owned. Rank by users
affected. An error with a stack trace is already a reproduction lead.
- CI on main:
pnpm ci:red-report (exit 2 means CI unavailable, not
clean).
- Chat run health:
node scripts/chat-health.mjs --hours 72; see
chat-run-health.md for what each failure
reason usually means.
- Deploys and releases:
gh run list --branch main --status failure --limit 30
for the deploy, release, desktop, and publish workflows.
2. Triage: lean toward fixing
Put each report in one bucket:
- Fix: anything objectively broken. This means errors, nothing happening
on click, lost or wrong data, regressions, slowness, jank, and visible
defects like clipped or overlapping text, unreadable contrast, or a broken
layout. A credible "nothing happens" counts as evidence.
- Build: a feature request or UX change carrying a maintainer's
:upvote: or :yes-party:.
- Needs a person (🎫): subjective product or UX changes no maintainer has
endorsed (add, move, hide, or restyle controls; "hard to find"), work that needs access or settings only
a person has, a provider outage, or a report you can't act on without one
missing fact. A bug you couldn't reproduce by reading code is still a Fix.
- Not actionable: praise, chatter, announcements, things a human already
answered, duplicates of something already in this run. These get no
reaction and no reply.
Issues from the error and health sources above go in the same buckets,
ranked with the Slack reports by people affected. They have no thread to react on,
so they appear only in the recap.
Claim Fix reports in order of people affected. Unfinished claims stay in
"Claimed reports" for the next run. Any Fix, Build, or Needs-a-person root you leave
unclaimed gets no reaction, only one line in the recap saying why, so
nothing is dropped silently. That covers every candidate your source helpers
return: each one ends the run either claimed or on a recap line with its
reason.
In scheduled and manual runs, record each root (channel and ts) under "Claimed
reports" in the feedback automation's memory.md, then add and verify 👀
before investigating. A record alone does not establish a claim: the root
must also have the verified 👀 from this workflow. Retain the record until
the report is resolved or a person takes over. If shared memory is unavailable,
do not claim new roots.
3. Fix as many as you can
- Group related reports so one cause gets one fix. Give each independent
group a fresh subagent on your own model and effort, with disjoint files,
or fix it in the main thread. Never hand a finished helper a new job; start
a fresh one. The main thread plans, reviews diffs, and integrates.
- Reproduce before you edit. Get the failure on screen first: a test
that fails, the action returning the wrong thing, a stack trace, or the
flow broken in a signed-in local app. Reading code is not an attempt:
write the report's steps as a failing test, or walk them in the signed-in
app. A broken setup isn't an attempt either: when a test can't load or an
install fails, fix the setup (install what's missing, or free space by
removing only this run's own build output) or
walk the steps in the signed-in app instead. Only after a real attempt fails does a report become 🎫, and its
recap line says what you ran. No reproduction means no code change.
- Check the history.
git log --since=30.days --oneline -- <files>: if
this surface was fixed before for the same symptom, that fix treated a
symptom. Find the shared cause (fix-at-the-boundary) instead of adding
one more patch, and look for the same pattern elsewhere.
- Prove it. The check has to fail with the fix's source changes put
back to
main and pass with them. A check that passes without your fix
doesn't test the bug: write one that does. Tests that read source text
never count. For a purely visual fix, before and after screenshots are
evidence, not automated proof: before merging, a human must open both and
confirm they show the reported change. Put the evidence under the report in
the PR's Reports section. A test-only change (test files, optional
.changeset/) needs only fix-proof: test-only.
Commit the fix and its check, then run
node scripts/fix-proof.ts --fix <that commit> -- <the check> from the
repo root and paste its fix-proof: proven ... line, unchanged, under
each report it fixes (visual: <before> <after> for a purely visual
fix). The required Lint & format check runs this gate for PRs with a
Reports section and re-runs machine proofs. It marks visual-only reports
as needing human validation before merge. The second
opinion below reads the red output to confirm the check failed on the
bug, not on setup.
- Two misses, then stop guessing. If two attempts don't turn the check
green, stop editing and capture the real failure first (stack, logs, the
reporter's actual data). If that doesn't settle it within about 30
minutes, it becomes 🎫 with one line on what you learned.
- Second opinion before the PR, every time. Give a fresh subagent on
your own model only the report text, the diff, and the proof, and ask: does
the check exercise what the reporter saw, is this the cause or a symptom,
and what edge case breaks it? Fix what it finds before you open the PR.
- Keep fixes small. Each report's fix goes in its own commit so a bad
one can be reverted alone. A fix that grows past about 300 lines, or into
a new subsystem (timeouts, cancellation, retries) the reporter didn't ask
for, isn't a sweep fix: ship the small part and 🎫 the rest. Match the
versions the workspace already pins; a sweep is not a dependency upgrade.
- Never tune agent instructions or prompts around one report's wording.
4. Ship
Put every fix in one PR, and open it once every claimed Fix report has a
fix or a failed attempt, so CI waits come after the fixing. Don't open one
PR per item. End the PR body with a
## Reports heading listing the link of every report the PR fixes: the
Slack thread, or the Sentry, Analytics, or GitHub issue. That list is the
only set of threads a later /ship may close out as fixed.
Wait for CI with one command that blocks until every check finishes, and
wait on it with the longest call your tool allows, not a short call you
repeat: every call re-sends your whole context.
Here that command is pnpm ci:wait <number>: it prints nothing until CI is
done, then one summary. Exit 1 means a check failed. Exit 2 means it couldn't
tell, and its last line says why: rerun a cancelled check, run the command
again when time ran out with checks still going, and otherwise look at the PR
yourself. To need fewer calls, pass --max-minutes about a minute
under your tool's call limit.
Without a repo command for it, run
gh pr checks <number> --watch --interval 60 > /dev/null in one call bounded
by your tool's limit, then read gh pr checks <number> twice, a minute
apart: CI is done only when both reads list the same checks, at least one,
and none is pending. Otherwise wait again the same way and rerun any cancelled check; after
about two hours, stop and look at the PR yourself.
Fix each review round's findings in one batch. If a second round still finds
a problem in the same fix, take that fix out of the PR, remove its whole entry
from the PR's Reports section: the link and its fix-proof: and visual:
lines (close-out marks every listed report fixed, and a stray evidence line
counts toward the report above it). Then 🎫 its
report with what review found, and merge the rest. Close the PR only when nothing verified
is left in it.
Merge it the way this repo merges PRs. In a Factory workspace, that means
the workflow's merge step after its merge gate, never a direct merge.
In the framework repo, run /ship through the merge.
A sweep's PR contains product fixes only. Don't edit this skill, add checker
scripts, or add bookkeeping files in it. If the run taught something, propose
a one-line lesson in the recap for a maintainer to accept or drop.
5. Close the loop
This step is the only place a claimed report gets its closing reply. Run it
after the merge for the threads in the PR's Reports section. If the merge
happened outside this run, the next sweep runs it: when you collect, list the
sweep PRs merged in the window (gh pr list --state merged --limit 100 --search "merged:>=<date>") and close out every report in their Reports sections
that still lacks its ✅, however old the report is.
Entries that aren't Slack threads get no Slack action: GitHub issues follow
the issue rule below, and Sentry and Analytics issues are only listed in the
recap.
In the framework repo, /ship step 6 runs this step after its merge, and
address-feedback-with-replies runs it for its own threads.
Run it once per thread: if a reply in the thread already links this PR, just make sure ✅ is
there and move on.
Write only as the connected Slack profile, only in the feedback
channels, and in the reply format from the contract above. First re-read
the contract and this step from fresh origin/main (git fetch origin main && git show origin/main:.agents/skills/review-latest-feedback/SKILL.md): a
run that started hours ago may be holding older rules. Re-read each thread before
writing. If a maintainer has since added :mag: or a person has taken it over,
leave it. If a newer human reply says it's still broken, leave the 👀 for the
next sweep. Otherwise:
- Merged fix: add ✅ and reply that it's fixed, with the PR linked, and
will be on beta shortly. Only for reports whose
Reports entry carries a
proof (step 3). Without one, it's 🎫 and a reply that the fix is logged for a
person to check.
- Fix ready but not merged yet (for example CI is still running when the
run must end): reply that a fix (PR linked) is in review. Don't mention
beta yet. The next run adds ✅ once it's merged, with no second reply.
- Needs a person: add 🎫 and say in plain words what's known and what's
needed, without naming anyone ("thanks - logged for a product decision").
- Couldn't reproduce: add 🎫. If one small answer from the reporter would
unblock it and you couldn't find it yourself, ask for just that. Otherwise
say you couldn't reproduce it yet. Never ask
for anything already in the thread.
GitHub issues fixed by the merged PR get a one-line thank-you comment linking
the PR, and then get closed. Post it even when someone else opened the issue,
or the PR already closed it: reporters should be told.
Report unresolved CI, deploy, and chat-health failures in the private run
recap. Do not create Slack channel messages or GitHub issues for them.
Before ending, re-read each thread you touched and check:
- every 👀 from this run has an investigated disposition and, unless a person
has taken over, a reply plus ✅, 🎫, or an open fix PR listed under
Reports;
- every ✅ corresponds to a merged fix;
- no reply is technical or duplicated.
6. Recap
Keep the recap short, for the maintainer reading it:
## Feedback sweep - <date, Pacific>
Fixed and merged (N): <one line each, with Slack thread link> · PR <link>
Closed out from earlier runs (N): <Slack thread link> · PR <link>
Needs a person (N): <what's needed, who, what you tried>
Still open from earlier runs (N): <why>
Seen, not picked (N): <Slack thread link> - <one-line reason>
Roots: <n> seen, <n> claimed, <n> left
Sources unavailable: <list or "none">
Lesson (optional): <one line>
"Fixed and merged" is this run's PR only; reports you closed out for an
earlier run's merged PR go under "Closed out from earlier runs".
In an automation, append the same recap to its memory.md.
1---2name: review-latest-feedback3description: >-4 Sweep recent feedback (the Slack feedback channels, GitHub issues, CI on5 main, Sentry, Analytics errors, chat run health, deploys), fix as many real6 bugs as possible, ship them in one PR, then close the loop on every Slack7 report with 👀/✅/🎫 and one short plain-English reply. Use for scheduled or8 manual feedback sweeps, usually followed by /ship.9user-invocable: true10scope: dev11metadata:12 source: https://skillmd.com/skills/builderio/review-latest-feedback13 internal: true14---1516# Review Latest Feedback1718The job: find what's broken, fix as much of it as you can, ship it, and tell19each reporter in one friendly sentence what happened. A sweep is judged by20bugs fixed and merged, and by every claimed report getting a clear answer. It is21not judged by ledgers, receipts, or checker output.2223## The contract2425The detailed Slack reaction and reply rules are in26[slack-reaction-replies.md](references/slack-reaction-replies.md); read it before27marking, replying to, or closing any Slack report.2829Slack reactions on a report's root message mean exactly this:3031| Mark | Meaning | When to add |32| ---- | ----------------------------------------- | ------------------------------------------------------------------------------------------------ |33| 👀 | "I'm on this" | Before you investigate, and only on reports you will finish answering in this run |34| ✅ | "Fixed" | After the fix is merged to `main` (main deploys to beta automatically) |35| 🎫 | "Needs a person" | A product call, access or credentials, info from the reporter, or a fix you couldn't land |3637A maintainer is a Builder.io team member in the workspace who is not the38reporter and not a bot. Maintainers steer the sweep with reactions on a root:3940| A maintainer adds | Meaning |41| ----------------- | ----------------------------------------------------------------------------------------- |42| `:mag:` | Don't work on this; someone else will. Wins over everything below |43| `:yes-party:` | Definitely do this, even if it looks ambiguous or like a UX change. Build the smallest version that does what was asked |44| `:upvote:` | Same as `:yes-party:` for a feature request |45| 👀 by hand | They're working on it themselves; leave it (see "Open items first" for how to tell their 👀 from ours) |4647Reactions can't be removed through the Slack connector, so never add one you48might need to take back. Every 👀 you add ends the run with a reply, plus49either ✅ or 🎫, or a fix PR that is still open and lists the report (its50merge adds the ✅).5152Every claimed report gets **one** reply in its thread per outcome. The one exception is53the merged `✅` close-out after an in-review reply, which gets no second reply. Each reply must be:5455- **An answer first:** if they asked something ("is this expected?"),56 answer it in the first words ("no, that's a bug"), not with a status.57- **Short:** one sentence, about 25 words at most, for someone who doesn't58 read code.59- **About what they'll notice:** say what changes for them ("recordings no60 longer get stuck on Saving"), not how it works ("cleanup now waits for the61 transcript"). Leave out file, function, and service names, PR numbers,62 ticket IDs, CI, logs, runs, "repro", "verified", "scope", "sweep", status63 labels ("In progress:", "Skipped:", "Resolved elsewhere:", "Blocked on64 human owner:", "Open - no question:"), and any mention of this workflow.65 Never report your own progress ("unreproduced", "not traced yet",66 "remains open"); say what happens next instead.67- **Clear on what happens next:** "I'll ..." when this workflow acts. When it68 needs a person, never name one and never say "the team", "product", "the69 framework team", "the owner", or "someone": say only what's known ("I70 couldn't reproduce this yet", "logged for a product decision"), add 🎫, and71 record it as needing a person with the reason. Never @-mention anyone.72- **Not a to-do list for them:** look things up yourself first (the73 message's time and author, their account and recent activity, logs, the74 linked page or screenshot). Ask only when nothing else can unblock it, and75 then ask one small thing, in this one reply. Never send a list of76 questions, never ask them to test, retry, re-record, pull, reinstall, or77 gather logs, and never come back later asking for something you could have78 asked the first time. If it needs access you don't have (credentials, admin79 rights, account changes), say it's logged for someone with access and add80 🎫; never hand it to the reporter.81- **Plain, not preachy:** don't explain how the product works unless they82 asked, don't tell them what they should do, and skip apologies, hedging,83 "actually", "just", "simply", and "Great catch!!".84- **True:** "fixed" only for a merged fix, and "on beta shortly" only after85 the merge. Before that, the fix is "in review". Never "fixed and verified"86 on the strength of a local check.87- **Once:** never post the same reply twice, and never follow up with one88 that adds nothing new.89- Link the PR on the word "fix" (`<url|fix>`), never by number.90- End with the bold line `**This was sent from a bot**` on its own line,91 once.9293Don't reply at all to jokes, venting, banter, a message meant for a specific94teammate, or a thread a person is already handling. Once a person says they95have it, or asks the bot to stop, post nothing more in that thread. Before posting, reread the reply as the reporter: would they get it96in five seconds, know what happens next, and feel we did the legwork? Check97the draft before sending, not after: the first version is the one that98notifies them, so editing it afterwards doesn't undo it.99100Good (each is the first line; the footer goes on the line below it):101102- `thanks - the <https://github.com/BuilderIO/agent-native/pull/1234|fix> so the export keeps your column order is merged and will be on beta shortly.`103- `thanks - once this <https://github.com/BuilderIO/agent-native/pull/1234|fix> merges, the export will keep your column order; it's in review now.`104- `thanks - where that button should go is logged for a product decision.`105- `thanks - I couldn't find which deck this was; could you paste its link?`106107So a full reply looks like:108109```text110thanks - the <https://github.com/BuilderIO/agent-native/pull/1234|fix> so the export keeps your column order is merged and will be on beta shortly.111**This was sent from a bot**112```113114Real replies that missed, and what to send instead:115116| Sent | Instead |117| --- | --- |118| "Could you send the recording link, approximate time, and browser so we can trace whether the clip produced media before it tried to save?" | Find the recording from the reporter and message time first. If you still can't: "thanks - I couldn't find which recording this was; could you paste its link?" |119| "cleanup now waits for the transcript, and a fix is ready. It should be on beta shortly." (sent before the merge) | "thanks - recordings will keep their transcript once this <url\|fix> is merged; it's in review now." |120| "I've left moving it somewhere easier to find for product to decide." | "thanks - where the editor should live is logged for a product decision." |121| "Partially fixed - PR #7512 merged the project-instructions/context carryover fix", after she'd asked whether FullStory could be added for the whole org | Answer that question first: "yes - I'll add it for everyone" or "not yet - that's logged for a decision". |122| "All fixed and verified. Please try again!" plus instructions to pull main and reinstall | "thanks - this <url\|fix> is merged and will be on beta shortly." No homework. |123| More replies after Shawn wrote "bot please stop! I've got this all fixed and in flight" | Nothing. A person has it. |124| "Thanks for correcting me, Julia - I misread the replies. ... Open - no question: I don't have enough source evidence to claim a fix." | Nothing: the thread was a joke. |125| Asked for the app URL, what was clicked, and the time; then, in a second reply, for the time zone. | One reply, one question, only for what you can't look up. The message's own timestamp already says when. |126127## Boundaries128129- **Slack:** only `#product-agent-native-feedback` (`C0ATH3CCZT4`)130 and `#qa-agent-native` (`C0C4U4XRT6X`), a narrower set the invocation131 names, or a caller's list (as in `factory-run`). Never132 react or reply elsewhere. Reply only in a report thread this workflow133 claimed with 👀 and investigated; a bot reply alone isn't a claim.134 Never start a new channel message.135- **Leave a report alone** when:136 - a maintainer put `:mag:` or their own 👀 on it (table above);137 - someone in the thread asked a specific person to look, a teammate says138 they're handling it, or a maintainer asked for a human on it;139 - another person's 👀 sits on it and they've been active in the last 3140 days;141 - an earlier run's PR for it is still open. That PR's `/ship` closes the142 report out when it merges.143- **PRs:** push to, comment on, or merge only the PR this run opened. PRs from144 other sessions share the active GitHub login (`gh api user --jq .login`), so145 authorship proves nothing: a PR under that login may be another thread's146 work in flight, and pushing to it lands unreviewed commits on it. Never touch any other PR unless the invocation names that exact PR and147 action. That holds inside `/ship` too: its "Your PRs" means this run's PR.148- **One run, one PR.** Every sweep PR is authored by the active GitHub login, so the149 author check can't tell runs apart: an150 open PR with a `Reports` section that this run didn't open belongs to151 another run that may still be working on it. Don't push to, reply on, or152 merge it; list it under "Still open from earlier runs".153- **Goals end with the run.** If you set a Codex goal, scope it to this run:154 name its PR once you have one, and end it after step 6 ("merge PR #N,155 close its Reports, append the recap"). Never set an open-ended goal such as156 "review feedback" or "fix my open PRs". Goals resume after restarts,157 so an open-ended one keeps running into the next scheduled sweep and158 collides with it.159- Use the connected Slack integration for every Slack read and write. Confirm160 which profile is connected before the first write; every reply and reaction161 posts as that profile.162163## Dry run164165On "dry run", run every step but send nothing:166167- Record each reaction and the exact reply you'd post, linking the fix branch168 where the PR would go.169- Treat other runs' claims as unclaimed; honor a maintainer's `:mag:` and 👀.170- Publish one branch per fix group; skip PRs, merges, `/ship`, and comments.171- Skip the connected-profile check; use the state the invocation names.172173<!-- framework-repo-only:start -->174The factory VM runs this skill through `factory-run`, which differs on175purpose: replies post as the factory's own Slack app (no bot line, no profile176check), claims live in factory state, each fix group gets its own branch and177PR, and the watcher merges.178<!-- framework-repo-only:end -->179180## 1. Collect (keep it under ~15 minutes)181182Work in a task-owned worktree on fresh `origin/main`. If you start in a shared183checkout, or the expected worktree is missing, create one184(`git worktree add <path> origin/main`) and carry on. That is never a reason185to stop.186187Window: roots posted in the last 5 days (or since the timestamp or permalink188the invocation gives), for new reports and open items alike. An older thread189is in the window too when we have already replied in it and a person has190replied since the window began: a channel read only shows roots, so search for191recent thread replies (`in:<#channel> is:thread after:<date>`) to find them.192Roots in `memory.md` under "Claimed reports", "Claimed, not yet answered",193or "Must handle next run" are always in the window. Other older threads are left as they are, except reports listed under194`Reports` in a sweep PR opened or merged in that window: those always get195their close-out.196197Give each source below its own helper on the cheapest model (Luna in Codex,198Haiku in Claude Code), and close it as soon as it reports. A helper gathers;199you triage. The Slack helper returns every root a person posted in the window with no ✅, 🎫, or `:mag:`,200and every root with a person's reply newer than ours, one line each (link,201author, first line). It never filters, ranks, or drops one: a new report has202no reaction and no thread yet.203204- **Read the channels.** Use `slack_read_channel` with205 `response_format: detailed` (paginate) back 5 days. Its output shows206 reaction counts but not who added them, so call `slack_get_reactions` on any207 root showing `eyes`, `white_check_mark`, `ticket`, `mag`, `yes-party`, or208 `upvote`. Avoid `hasmy:`; run the209 [backstop](references/slack-reaction-replies.md#backstop-for-unclaimed-roots).210- **Open items first.** Verify the connected profile's 👀 on the root and workflow ownership211 from a matching run recap, proof-bearing PR, or automation memory claim.212 A bot reply, reaction, or memory entry alone is insufficient. Investigate213 before replying, and read the thread for human updates. A root214 listed in the `Reports` section of a sweep215 PR that is still open and that this run didn't open belongs to the run216 working on that PR: don't react217 or reply there, and leave it out of your self-check; just list it under218 "Still open from earlier runs". If that PR has had no push in 24 hours,219 its run has stopped: its reports are yours again. Of the remaining220 workflow roots, those without ✅ or 🎫 come first in this run. A newer221 human reply is context, not by itself a reason to respond: investigate the222 claimed report and leave conversations already handled by people alone. A223 root with 🎫 and our reply after its latest human message is already answered: you may still fix it, but post nothing there224 until the fix merges. A225 reply like "still broken" after a ✅ means the earlier fix failed. Find what226 that fix changed and why it didn't hold before trying again.227- **New reports:** roots from the last 5 days. React and reply on the root228 message's own `ts`. A channel-history row whose `thread_ts` differs from its229 `ts` is a reply that was also sent to the channel: use its `thread_ts`230 parent instead. Read the thread for every root231 that looks like a report. Open screenshots and attachments, since the error232 text in a screenshot is often the whole diagnosis.233- **Other sources.** Unread or failing sources are reported as234 **unavailable**, never as "nothing found".235 - GitHub: open issues updated in the window236 (`gh issue list --state open --search "updated:>=<date>"`). Skip ones237 someone else has offered to fix.238 - Sentry (through its connector, or the local reader the invocation239 names) and Agent-Native Analytics (`list-error-issues`, then240 `get-error-issue`): unresolved, recent, and repo-owned. Rank by users241 affected. An error with a stack trace is already a reproduction lead.242<!-- framework-repo-only:start -->243 - CI on main: `pnpm ci:red-report` (exit 2 means CI unavailable, not244 clean).245 - Chat run health: `node scripts/chat-health.mjs --hours 72`; see246 [chat-run-health.md](references/chat-run-health.md) for what each failure247 reason usually means.248 - Deploys and releases: `gh run list --branch main --status failure --limit 30`249 for the deploy, release, desktop, and publish workflows.250<!-- framework-repo-only:end -->251252## 2. Triage: lean toward fixing253254Put each report in one bucket:255256- **Fix:** anything objectively broken. This means errors, nothing happening257 on click, lost or wrong data, regressions, slowness, jank, and visible258 defects like clipped or overlapping text, unreadable contrast, or a broken259 layout. A credible "nothing happens" counts as evidence.260- **Build:** a feature request or UX change carrying a maintainer's261 `:upvote:` or `:yes-party:`.262- **Needs a person (🎫):** subjective product or UX changes no maintainer has263 endorsed (add, move, hide, or restyle controls; "hard to find"), work that needs access or settings only264 a person has, a provider outage, or a report you can't act on without one265 missing fact. A bug you couldn't reproduce by reading code is still a Fix.266- **Not actionable:** praise, chatter, announcements, things a human already267 answered, duplicates of something already in this run. These get no268 reaction and no reply.269270Issues from the error and health sources above go in the same buckets,271ranked with the Slack reports by people affected. They have no thread to react on,272so they appear only in the recap.273274Claim Fix reports in order of people affected. Unfinished claims stay in275"Claimed reports" for the next run. Any Fix, Build, or Needs-a-person root you leave276unclaimed gets no reaction, only one line in the recap saying why, so277nothing is dropped silently. That covers every candidate your source helpers278return: each one ends the run either claimed or on a recap line with its279reason.280281In scheduled and manual runs, record each root (channel and ts) under "Claimed282reports" in the feedback automation's `memory.md`, then add and verify 👀283before investigating. A record alone does not establish a claim: the root284must also have the verified 👀 from this workflow. Retain the record until285the report is resolved or a person takes over. If shared memory is unavailable,286do not claim new roots.287288## 3. Fix as many as you can289290- Group related reports so one cause gets one fix. Give each independent291 group a fresh subagent on your own model and effort, with disjoint files,292 or fix it in the main thread. Never hand a finished helper a new job; start293 a fresh one. The main thread plans, reviews diffs, and integrates.294- **Reproduce before you edit.** Get the failure on screen first: a test295 that fails, the action returning the wrong thing, a stack trace, or the296 flow broken in a signed-in local app. Reading code is not an attempt:297 write the report's steps as a failing test, or walk them in the signed-in298 app. A broken setup isn't an attempt either: when a test can't load or an299 install fails, fix the setup (install what's missing, or free space by300 removing only this run's own build output) or301 walk the steps in the signed-in app instead. Only after a real attempt fails does a report become 🎫, and its302 recap line says what you ran. No reproduction means no code change.303- **Check the history.** `git log --since=30.days --oneline -- <files>`: if304 this surface was fixed before for the same symptom, that fix treated a305 symptom. Find the shared cause (`fix-at-the-boundary`) instead of adding306 one more patch, and look for the same pattern elsewhere.307- **Prove it.** The check has to fail with the fix's source changes put308 back to `main` and pass with them. A check that passes without your fix309 doesn't test the bug: write one that does. Tests that read source text310 never count. For a purely visual fix, before and after screenshots are311 evidence, not automated proof: before merging, a human must open both and312 confirm they show the reported change. Put the evidence under the report in313 the PR's `Reports` section. A test-only change (test files, optional314 `.changeset/`) needs only `fix-proof: test-only`.315<!-- framework-repo-only:start -->316 Commit the fix and its check, then run317 `node scripts/fix-proof.ts --fix <that commit> -- <the check>` from the318 repo root and paste its `fix-proof: proven ...` line, unchanged, under319 each report it fixes (`visual: <before> <after>` for a purely visual320 fix). The required `Lint & format` check runs this gate for PRs with a321 `Reports` section and re-runs machine proofs. It marks visual-only reports322 as needing human validation before merge. The second323 opinion below reads the red output to confirm the check failed on the324 bug, not on setup.325<!-- framework-repo-only:end -->326- **Two misses, then stop guessing.** If two attempts don't turn the check327 green, stop editing and capture the real failure first (stack, logs, the328 reporter's actual data). If that doesn't settle it within about 30329 minutes, it becomes 🎫 with one line on what you learned.330- **Second opinion before the PR, every time.** Give a fresh subagent on331 your own model only the report text, the diff, and the proof, and ask: does332 the check exercise what the reporter saw, is this the cause or a symptom,333 and what edge case breaks it? Fix what it finds before you open the PR.334- **Keep fixes small.** Each report's fix goes in its own commit so a bad335 one can be reverted alone. A fix that grows past about 300 lines, or into336 a new subsystem (timeouts, cancellation, retries) the reporter didn't ask337 for, isn't a sweep fix: ship the small part and 🎫 the rest. Match the338 versions the workspace already pins; a sweep is not a dependency upgrade.339- Never tune agent instructions or prompts around one report's wording.340341## 4. Ship342343Put every fix in one PR, and open it once every claimed Fix report has a344fix or a failed attempt, so CI waits come after the fixing. Don't open one345PR per item. End the PR body with a346`## Reports` heading listing the link of every report the PR fixes: the347Slack thread, or the Sentry, Analytics, or GitHub issue. That list is the348only set of threads a later `/ship` may close out as fixed.349350Wait for CI with one command that blocks until every check finishes, and351wait on it with the longest call your tool allows, not a short call you352repeat: every call re-sends your whole context.353<!-- framework-repo-only:start -->354Here that command is `pnpm ci:wait <number>`: it prints nothing until CI is355done, then one summary. Exit 1 means a check failed. Exit 2 means it couldn't356tell, and its last line says why: rerun a cancelled check, run the command357again when time ran out with checks still going, and otherwise look at the PR358yourself. To need fewer calls, pass `--max-minutes` about a minute359under your tool's call limit.360<!-- framework-repo-only:end -->361Without a repo command for it, run362`gh pr checks <number> --watch --interval 60 > /dev/null` in one call bounded363by your tool's limit, then read `gh pr checks <number>` twice, a minute364apart: CI is done only when both reads list the same checks, at least one,365and none is pending. Otherwise wait again the same way and rerun any cancelled check; after366about two hours, stop and look at the PR yourself.367368Fix each review round's findings in one batch. If a second round still finds369a problem in the same fix, take that fix out of the PR, remove its whole entry370from the PR's `Reports` section: the link and its `fix-proof:` and `visual:`371lines (close-out marks every listed report fixed, and a stray evidence line372counts toward the report above it). Then 🎫 its373report with what review found, and merge the rest. Close the PR only when nothing verified374is left in it.375376Merge it the way this repo merges PRs. In a Factory workspace, that means377the workflow's merge step after its merge gate, never a direct merge.378<!-- framework-repo-only:start -->379In the framework repo, run `/ship` through the merge.380<!-- framework-repo-only:end -->381382A sweep's PR contains product fixes only. Don't edit this skill, add checker383scripts, or add bookkeeping files in it. If the run taught something, propose384a one-line lesson in the recap for a maintainer to accept or drop.385386## 5. Close the loop387388This step is the only place a claimed report gets its closing reply. Run it389after the merge for the threads in the PR's `Reports` section. If the merge390happened outside this run, the next sweep runs it: when you collect, list the391sweep PRs merged in the window (`gh pr list --state merged --limit 100 --search392"merged:>=<date>"`) and close out every report in their `Reports` sections393that still lacks its ✅, however old the report is.394Entries that aren't Slack threads get no Slack action: GitHub issues follow395the issue rule below, and Sentry and Analytics issues are only listed in the396recap.397<!-- framework-repo-only:start -->398In the framework repo, `/ship` step 6 runs this step after its merge, and399`address-feedback-with-replies` runs it for its own threads.400<!-- framework-repo-only:end -->401Run it once per thread: if a reply in the thread already links this PR, just make sure ✅ is402there and move on.403404Write only as the connected Slack profile, only in the feedback405channels, and in the reply format from the contract above. First re-read406the contract and this step from fresh `origin/main` (`git fetch origin main407&& git show origin/main:.agents/skills/review-latest-feedback/SKILL.md`): a408run that started hours ago may be holding older rules. Re-read each thread before409writing. If a maintainer has since added `:mag:` or a person has taken it over,410leave it. If a newer human reply says it's still broken, leave the 👀 for the411next sweep. Otherwise:412413- **Merged fix:** add ✅ and reply that it's fixed, with the PR linked, and414 will be on beta shortly. Only for reports whose `Reports` entry carries a415 proof (step 3). Without one, it's 🎫 and a reply that the fix is logged for a416 person to check.417- **Fix ready but not merged yet** (for example CI is still running when the418 run must end): reply that a fix (PR linked) is in review. Don't mention419 beta yet. The next run adds ✅ once it's merged, with no second reply.420- **Needs a person:** add 🎫 and say in plain words what's known and what's421 needed, without naming anyone ("thanks - logged for a product decision").422- **Couldn't reproduce:** add 🎫. If one small answer from the reporter would423 unblock it and you couldn't find it yourself, ask for just that. Otherwise424 say you couldn't reproduce it yet. Never ask425 for anything already in the thread.426427GitHub issues fixed by the merged PR get a one-line thank-you comment linking428the PR, and then get closed. Post it even when someone else opened the issue,429or the PR already closed it: reporters should be told.430<!-- framework-repo-only:start -->431Report unresolved CI, deploy, and chat-health failures in the private run432recap. Do not create Slack channel messages or GitHub issues for them.433<!-- framework-repo-only:end -->434435Before ending, re-read each thread you touched and check:436437- every 👀 from this run has an investigated disposition and, unless a person438 has taken over, a reply plus ✅, 🎫, or an open fix PR listed under `Reports`;439- every ✅ corresponds to a merged fix;440- no reply is technical or duplicated.441442## 6. Recap443444Keep the recap short, for the maintainer reading it:445446```md447## Feedback sweep - <date, Pacific>448Fixed and merged (N): <one line each, with Slack thread link> · PR <link>449Closed out from earlier runs (N): <Slack thread link> · PR <link>450Needs a person (N): <what's needed, who, what you tried>451Still open from earlier runs (N): <why>452Seen, not picked (N): <Slack thread link> - <one-line reason>453Roots: <n> seen, <n> claimed, <n> left454Sources unavailable: <list or "none">455Lesson (optional): <one line>456```457458"Fixed and merged" is this run's PR only; reports you closed out for an459earlier run's merged PR go under "Closed out from earlier runs".460461In an automation, append the same recap to its `memory.md`.