Skip to content

🐛 Filter em-dashes on every outbound path, agent proxy writes included - #19110

Merged
hivecommons-hive[bot] merged 8 commits into
v5from
fix/emdash-outbound-durable
Nov 19, 2026
Merged

hivecommons-hive[bot] merged 8 commits into
v5from
fix/emdash-outbound-durable

Conversation

@clubanderson

@clubanderson clubanderson commented Nov 18, 2026 •

Copy link
Copy Markdown
Member

Fixes #19103

Root cause

#18970 added the shared dash filter in src/pkg/prosefilter and applied it to hive's own GitHub writes. Agent writes made with gh api go through the MITM egress proxy, and proxyHTTPHost in src/crux/proxy/github_proxy_mitm.go forwarded an allowed write with req.Write(upstream) (the forward at the end of the request loop) without ever touching the body. The proxy lives in the nested crux module, which cannot import pkg/, so no proxy path could call the filter.

What changes

  • One filter, moved to crux. src/crux/prosefilter now holds the implementation (same replacement rule as ✨ Stop hive from writing em-dashes: shared outbound filter, ASCII trailer, base-prompt rule, ratchet test #18970, code fences, inline code and blockquotes untouched) plus SanitizeJSON, which applies it to string values under named JSON keys at any depth. src/pkg/prosefilter only delegates.
  • Proxy. sanitizeAgentWriteProse (src/crux/proxy/outbound_prose.go) runs after every gate, just before forwarding. It rewrites REST body/title/message (review comments[].body included), the body/title variables of comment, review, review-thread reply, PR and issue GraphQL mutations, and Linear mutation body/title/description. It never touches non-JSON bodies, non-JSON content types, other fields, the GraphQL query text, git smart-HTTP or LFS traffic, or hive control-plane requests. Content-Length is recomputed; oversized bodies (over 1 MiB) are forwarded unchanged. One debug line with counts, never text.
  • Commit messages. The push broker rewrites the outgoing HEAD commit message with the same filter (src/crux/pushbroker/commit_message.go), using --cleanup=verbatim so only dashes change. Best-effort: a read or amend failure is logged and the push goes ahead unchanged. Older commits in a multi-commit push are not rebased; the squash commit hive writes is filtered (MergePRVia), so what lands on the default branch is clean.
  • Every other unfiltered outbound path now calls the filter at its send boundary (table below).
  • Switch. HIVE_AGENT_PROSE_FILTER (default on) turns off rewriting of agent-authored text (proxy and commit message). Hive-composed text is always filtered, as in ✨ Stop hive from writing em-dashes: shared outbound filter, ASCII trailer, base-prompt rule, ratchet test #18970.
  • Prompt rule. Already present: ✨ Stop hive from writing em-dashes: shared outbound filter, ASCII trailer, base-prompt rule, ratchet test #18970 added the plain-dash rule to every kick header (addPlainDashRule). Unchanged.
  • CI guard. TestOutboundSitesInventory (src/pkg/prosefilter/outbound_sites_test.go) AST-scans src/pkg, src/cmd and src/crux for every function that sends an HTTP request with a body, in the style of the notification write-site inventory. Each must be classed filtered (calls the filter), via (a named filtered function upstream), template (source under the emit ratchet), or machine (with a reason). A new sender that is not listed, or a filtered site that stops calling the filter, fails CI. It also pins the proxy, push broker, relay, PR, merge and notification boundaries. The emit ratchet now also covers pkg/forge, pkg/feedbackbundle and pkg/fleetreport.

Outbound inventory

Path Boundary Before After
Agent relay requests (PR, issue, comment, review) relayBody, CreatePR, issue-request watcher filtered filtered
Hive-signed commits createCommitOnBranch filtered filtered
Hive-rendered GitHub comments and issues source templates emit ratchet emit ratchet
Agent GitHub writes via proxy (REST + GraphQL) sanitizeAgentWriteProse no filtered
Agent Linear writes via proxy same no filtered
Agent HEAD commit message push broker no filtered
Squash / merge commit message MergePRVia no filtered
ntfy, Slack, Discord webhook alerts notify.(*Notifier).Send no filtered
Hub Slack DMs hub.sendSlackDM no filtered
Discord bot messages, embeds, edits, threads, topic discord.(*Bot).do, setChannelTopic no filtered
Hub Discord webhooks postDiscordRequestWebhook, GitHubActivityFeed.postDiscord no filtered
Hub ntfy alert pushAuthAuditAlert no filtered
Meeting chat and spoken text (chat.post, say, say.prewarm) meetings.(*Client).post no filtered
Hub-filed GitHub issue titles postHubGitHubIssue no filtered
Hub-filed issue bodies, seen-log comments, affected-hives edits pkg/fleetreport templates no emit ratchet (the affected-hives block is parsed back by its separators, so it is not rewritten after rendering)
Linear advisory digest PostDigest no filtered before the skip-if-unchanged compare; footer marker now ASCII, legacy marker still matched
Linear / Jira work-source writes linearGraphQL, jiraSource.doJSON no filtered
Linear agent activity linearagent.doGraphQL no filtered
DIBS inbox ideas DibsDestination.Deliver no filtered
GitLab / Gitea / Forgejo writes pkg/forge no emit ratchet
Email replies no sender wired n/a n/a

Tests

  • Proxy table test: REST comment, PR body, review reply, PR title, review inline comments, contents-API commit message, fenced code block kept, en-dash range kept, non-text field, non-JSON content type, non-JSON body, git push, LFS, GET, switch off, control plane, GraphQL review-thread reply, GraphQL createPullRequest title, GraphQL query and unrelated mutation untouched, Linear description; Content-Length checked. Oversized body, read error, and an end-to-end relay test through proxyHTTP.
  • SanitizeJSON table test, switch test.
  • Push broker: rewrite, clean message untouched, switch off, amend and read failures do not block the push.
  • Discord and meetings send-boundary tests.

Follow-ups

  • GraphQL string literals inlined in the query text are not rewritten.
  • Non-HEAD commits in a multi-commit push keep their messages.
  • pkg/fleetreport still renders em-dash separators in the affected-hives block and seen log (held by the emit ratchet); moving them to hyphens needs a parser that accepts both forms for already-filed issues.

Coordination: PR #19107 (opened by the hive scanner agent) reached the same design independently: the filter moved into crux/prosefilter with pkg/prosefilter delegating, and agent writes filtered in the egress proxy. Its tests (comment, review and GraphQL, other fields untouched, switch off) are covered by the proxy table test here. #19107 is closed as superseded by this PR.

Kept to new files and small hooks. The only shared-file edits in src/crux/proxy are one field in github_proxy.go and one call before the git-path branch in github_proxy_mitm.go, away from the hunks in #19072 and #19007.

🤖 Generated with Claude Code

@kubestellar-prow kubestellar-prow Bot added the dco-signoff: yes Indicates the PR's author has signed the DCO. label Nov 18, 2026
@kubestellar-prow

Copy link
Copy Markdown
Contributor

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign nicholasjackson for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details Needs approval from an approver in each of these files:

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

@kubestellar-prow kubestellar-prow Bot added the size/XXL Denotes a PR that changes 1000+ lines, ignoring generated files. label Nov 18, 2026
@github-actions

github-actions Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Open PR overlap check

Base branch: v5

Conflicts are checked with every PR merged onto the current tip of v5, so a PR that only needs a rebase is not reported as conflicting with its siblings.

PR Title Relation Files
#19101 ✨ Hub: send the confirmation reply for mail sent to a hive with DKIM, behind the signing key (off by default) same files src/docs/env-vars.md
#19007 🐛 fix: strip hop-identity headers and table-test proxy relays same files src/crux/proxy/github_proxy_mitm.go
#19102 🐛 Let the agent that opened a PR reply on its review threads same files src/crux/proxy/github_proxy.go
src/crux/proxy/github_proxy_mitm.go
src/docs/github-write-surface.md

@github-actions github-actions Bot added conflicts-with-open-pr Open PR has a merge conflict with another open PR and removed conflicts-with-open-pr Open PR has a merge conflict with another open PR labels Nov 18, 2026
@hivecommons-hive

hivecommons-hive Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor.

@clubanderson this PR is held at stage/071-needs-author and the hive will not merge it yet.

Failing checks:

  • test (rest 3/4)

Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path.

Review follow-ups: 2 open of 2 (below review.backlog.min_severity: P2)

Non-blocking findings below the backlog filing line, listed here instead of as issues. Tick them off as they are fixed; set review.backlog.min_severity: P3 to file every finding.

  • P3 info style — Local sortedKeys helper duplicates existing idiom (src/crux/proxy/outbound_prose.go:189): sortedKeys(map[string]bool) is redefined here; the repo already has several per-package copies (e.g. src/pkg/compliance/posture.go, src/pkg/review/review.go generic). Fine within crux, which cannot i…
  • P3 low intent-alignment — Merge API commit_title/commit_message not filtered (src/crux/proxy/outbound_prose.go:51): restProseKeys covers body/title/message only; agent gh api merges send commit_title/commit_message unfiltered. Not listed in 🐛 Em-dashes still reach GitHub through agent writes on the egress proxy #19103, so backlog.

Review findings from adjudicator-3 (folded here because this head already carries its hive review):

Intent alignment for #19103: the diff delivers the acceptance criteria and the six-point durable fix. One low-severity, non-blocking gap.

  • Proxy rewrites REST body/title/message and GraphQL body/title variables (src/crux/proxy/outbound_prose.go), after the gates, with Content-Length recomputed and git, LFS, non-JSON and the off switch left untouched.
  • Push broker covers the HEAD commit message and documents the limit (item 4).
  • Send boundaries and the TestOutboundSitesInventory CI guard cover items 3 and 5.
  • The other outbound edits (Discord, meetings, Jira, DIBS, Linear) are the inventory item 3 asks for, so they are not scope creep.

Out-of-scope note: the merge API's commit_title and commit_message (PUT /pulls/{n}/merge) are not in restProseKeys, so an agent merging via gh api still sends dashes. The issue lists only message, so this is backlog, not a blocker.

I could not check out the tree, so I read the diff and fetched files through gh only. I did not run the tests.

Findings not on a changed line

  • src/crux/proxy/outbound_prose.go:51: Optional follow-up: add commit_title and commit_message here so agent PUT /pulls/{n}/merge writes are filtered too.

Review findings from adjudicator (folded here because this head already carries its hive review):

Correctness review at 39fa502 (in-scope for #19103: filter agent proxy writes on every outbound path)

One blocking finding, posted as an inline thread: JSON-escaped dashes (—/–) bypass the proxy filter. The rest of the proxy path I read looks sound: the body-limit restore, the Content-Length and chunked handling, the GraphQL variables-only rewrite, and the git/LFS exclusion.

Findings not on a changed line

  • src/crux/prosefilter/json.go:78: mayHoldDash documents "raw or as a JSON \u escape", but the second check (lines 78-79) searches for the literal bytes E2 80 94 / E2 80 93 again. That is the same test as the first check, not an escape test. Any client that serialises with ASCII escaping (python json.dumps default, jq -a, many curl scripts) sends \u2014. mayHoldDash returns false, so SanitizeJSON returns the body untouched and the dash reaches GitHub. That leaves the outbound path 🐛 Em-dashes still reach GitHub through agent writes on the egress proxy #19103 exists to close open. The test at src/crux/prosefilter/json_test.go:15 is named "escaped dash" but its input holds a raw em-dash, so it hides the gap. Once the pre-check passes, json.Decoder decodes the escape and the filter itself works.

    	lower := bytes.ToLower(body)
    	return bytes.Contains(lower, []byte(`\u2014`)) || bytes.Contains(lower, []byte(`\u2013`))
    

    Also change the test input to a real escape, e.g. `{"body":"a \u2014 b"}`, and add a proxy-level case in outbound_prose_test.go.

hive/awaiting-ci: rebased onto v5 (one citation-line conflict in session-taint.md, took v5); the earlier coverage failure is not re-triaged until CI re-runs. CI pending - sweep will merge when green.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: b02e6bc73

Evidence bundle `#19110c8d3d6c` sha256:620cc17c5d11e874c03e397ff9a8426d6f951909670c9ccfd87bca4cc5fd4b75 — download from the dashboard or `hivectl review evidence #19110`.

merged by Hive trusted-author auto-merge: author clubanderson holds role owner and GitHub permission admin on repo; CI green at 0157cec.

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Style review (head 39fa502): conventions are followed and the structure is clean, with one filter in crux, pkg delegating, an inventory CI guard, a changelog fragment and docs. One low-severity readability note is inline; no blockers.

-- hive: agent=adjudicator-4 backend=claude model=claude-sonnet-5-5 claude=2.1.327

}

// graphQLProseMutationRe matches any name in graphQLProseMutations as a whole
// word. A substring test is the right posture here, unlike for a permission

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Comment says 'A substring test is the right posture' but graphQLProseMutationRe is a whole-word match (\b(...)\b). Suggest rewording:

Suggested change
// word. A substring test is the right posture here, unlike for a permission
// word. Matching the name anywhere in the document is the right posture here, unlike for a permission

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Decision needed: @clubanderson should decide whether the GitHub egress proxy may rewrite agent request bodies after its gates, and the push broker amend the agent's HEAD commit message, to enforce the dash rule.

Routing, not a defect report: this PR changes proxy and push-broker behaviour (trust surface), so a person should accept it.

What I read: src/crux/proxy/outbound_prose.go (runs after every gate, JSON-only, skips git/LFS/non-JSON, 1 MiB cap forwards unchanged, Content-Length and chunking reset only when a field changed) and src/crux/pushbroker/commit_message.go plus its call site at pushbroker.go:218-232 (best-effort amend, HEAD re-read after). I found no correctness defect in those paths.

The decision: after this lands the proxy mutates what an agent sent to GitHub/Linear, and the broker changes the pushed HEAD sha. Both are reasonable for the rule in #19103, and both are switchable via HIVE_AGENT_PROSE_FILTER=false, but they are a policy choice on a trust boundary rather than an engineering detail.

-- hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.327

@hivecommons-hive

Copy link
Copy Markdown
Contributor

hive/scanner: the failing coverage check looks branch-local: baseline triage says MERGE_BASE (this branch is 13 commits behind v5, base check not red). Suggest merging v5 into the branch with a signed-off commit and re-checking before diagnosing further. Not pushing, since the branch is busy.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

@hivecommons-hive

Copy link
Copy Markdown
Contributor

Scanner: the only red check is coverage; the branch is 17 commits behind v5 (baseline triage: MERGE_BASE). This is a human-authored branch, so I am not pushing. Suggest merging v5 in (signed off) and re-checking coverage before diagnosing further.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

@github-actions github-actions Bot removed the conflicts-with-open-pr Open PR has a merge conflict with another open PR label Nov 18, 2026
@kubestellar-prow kubestellar-prow Bot added the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Nov 18, 2026
The #18970 filter covered hive's own GitHub writes but not text agents
send themselves through the egress proxy, so a gh api comment kept its
em-dashes. Move the one shared filter into the crux module so the proxy
uses it, rewrite JSON text fields of allowed agent writes (REST and the
comment, review, PR and issue GraphQL mutations), rewrite the HEAD commit
message in the push broker, and route notifications, Discord, meeting
chat and speech, hub posts, Linear, Jira and DIBS writes through it.

A new inventory test fails CI when an outbound HTTP send is added without
the filter or a stated reason.

Fixes #19103

Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
…best-effort

Hub self-report bodies carry the fleetreport affected-hives block, which is
parsed back by its separators, so only their titles are filtered. The Linear
digest filters before its skip-if-unchanged compare and its footer is now
plain ASCII (the old marker is still matched). The persona voice sample is
operator-only and stays out of the dashboard import ratchet. A failed
commit-message rewrite is logged and never blocks a push.

Signed-off-by: Andy Anderson <andy@clubanderson.com>
@hivecommons-hive
hivecommons-hive Bot force-pushed the fix/emdash-outbound-durable branch from 42fcc4e to b02e6bc Compare November 18, 2026 21:42
@kubestellar-prow kubestellar-prow Bot removed the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Nov 18, 2026

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Two required checks are red on head b02e6bc, and both come from this diff. I read the proxy hook, src/crux/proxy/outbound_prose.go and the push broker amend path against the PR body; the design holds up.

Blocking CI (verified in the job logs)

  • relative links and anchors resolve: DRIFT src/docs/design/session-taint.md:84 src/crux/pushbroker/pushbroker.go:244 -> lacks anchor(s) near cited range: FindSecret. Adding the sanitizeHeadCommitMessage hook in pushbroker.go shifted the cited line. Fix: python3 src/scripts/check-docs-citations.py --fix and commit the doc change.
  • coverage: crux/pushbroker: 98.3% (floor 99%). The new commit_message.go has error branches (os.CreateTemp, WriteString, Close) that the listed amend/read-failure tests may not reach. I did not map the uncovered lines, so check the per-function profile for rewriteHeadCommitMessage.

What I checked and found sound

  • The proxy filter runs after every gate has read the body as the agent sent it, only on POST/PATCH/PUT with an empty or JSON content type, skips git paths, and forwards bodies over 1 MiB unchanged rather than truncating them. Content-Length is reset and TransferEncoding cleared only when a field actually changed.
  • GraphQL handling rewrites only variables, and only when the query is a mutation naming a listed prose mutation, so query text is never altered.
  • Amend uses --only --cleanup=verbatim, so the agent's staged changes stay out of the commit and trailers survive.

Open question for the author, not a defect: the 1 MiB pass-through means an agent that sends an oversized body skips the filter silently. The debug-only logging gives an operator no signal for that case.

Both blockers are mechanical. Once they are green this looks correct to me; merge stays a human call.

-- hive: agent=reviewer-2 backend=claude model=claude-sonnet-5-5 claude=2.1.327

…on-taint citations

The commit-message rewrite staged the text through a temp file whose create, write and close failures cannot be driven from a test, and the missing-blank-line check was unreachable because an empty message has no dash. Passing the message as one -m argument removes those branches; the amend-failure path stays covered by the existing fail-open test and the success log is now asserted. The secret-gate citations in session-taint.md move to the shifted line.

Signed-off-by: Andy Anderson <andy@clubanderson.com>

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Decision needed: @clubanderson should decide whether the egress proxy may rewrite agent-authored request bodies after its gates ran, and whether the push broker may amend an agent's HEAD commit, as proposed here.

Routing, not a defect report. I read the proxy hook, outbound_prose.go, SanitizeJSON and the push broker change against the base tree. I found no verified bug in the proxy path, but this PR changes trust surfaces (egress proxy request handling, push broker commit rewriting), so a person should sign off.

What I checked and found sound:

  • sanitizeAgentWriteProse runs after every gate has read the body (github_proxy_mitm.go hook), skips git and LFS paths, non-JSON content types, the internal caller and bodies over 1 MiB, and fixes Content-Length and TransferEncoding only when a field changed.
  • SanitizeJSON leaves numbers exact (UseNumber) and returns the original bytes when nothing changed.
  • GraphQL handling filters only variables, never query text, and only for mutations in graphQLProseMutations.

What a maintainer should decide:

  1. The proxy now mutates agent-authored bodies that other gates previously saw verbatim. That is a policy choice about the proxy's role, and it is the same surface #19072 and #19007 touch.
  2. rewriteHeadCommitMessage (src/crux/pushbroker/commit_message.go) runs git commit --amend under gitEnv(), which sets GIT_AUTHOR_* and GIT_COMMITTER_* to the lane identity. Open question, not verified: if HEAD was authored by a different identity, the amend would reassign the author, and the sign-off guard near pushbroker.go:484 requires the lane's own Signed-off-by when both identities match the lane. Please confirm the guard cannot then reject a push that passed before, or that this matches what stripTrailingBlankLines already does.
  3. Size: about 2000 added lines across 39 files, with a lot of outbound sites touched at once. The TestOutboundSitesInventory guard is a good idea, but it is a lot for one review. Splitting proxy plus push broker from the notification and Discord hooks would make each half reviewable.

CI was still running when I read this (golangci-lint, build, several test shards).

-- hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.327

@hivecommons-hive

Copy link
Copy Markdown
Contributor

@clubanderson this PR is back in your court (stage/071-needs-author). Its commits are signed off as the hive App rather than as you, so a squash merge would misattribute them and the sign-off attribution check stays red. No hive agent can rewrite a contributor's commits — the fix is yours:

  • 1 commit is signed off as the hive App instead of you, so a squash merge would misattribute it — set git config user.email to your own address, then run git rebase --exec 'git commit --amend --no-edit --reset-author -s' HEAD~5 && git push --force-with-lease

Once the new head lands, CI and the review swarm re-run automatically and the PR returns to the normal path. Current head: f65da8c66dff.

@hivecommons-hive

Copy link
Copy Markdown
Contributor

Scanner triage: Sign-off survives the squash and native-queue-policy fail on stale-branch drift (baseline action MERGE_BASE, branch behind v5). Suggested fix for the worker: merge origin/v5 with sign-off and push. Not pushing: branch is busy.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

@hivecommons-hive

Copy link
Copy Markdown
Contributor

Scanner triage: the failing checks here (native-queue-policy, Sign-off survives the squash) are not on the default branch; the branch is behind v5 and the checks are stale. Suggested fix: merge the base branch into this branch (signed off) and let CI re-run. This is a human-authored branch, so I am not pushing.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

@hivecommons-hive

Copy link
Copy Markdown
Contributor

Scanner triage: baseline check says this branch is behind its base by 16 commits and the failing checks (native-queue-policy, Sign-off survives the squash) are not shared with the base branch. Suggested fix: merge the base branch into this branch (signed off, git merge -s ort origin/<base> && git commit -s if needed) and re-run CI before diagnosing further. I am not pushing to this branch because it is not hive-authored.


🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

…bound-sites inventory

It replays the model prompt to the inference backend, like its sibling
retryInferenceOnAuthFailure, so it is a machine send site, not prose.

Signed-off-by: Andy Anderson <andy@clubanderson.com>
@clubanderson
clubanderson force-pushed the fix/emdash-outbound-durable branch from f65da8c to c8d3d6c Compare November 19, 2026 01:24
@hivecommons-hive
hivecommons-hive Bot merged commit f7daf99 into v5 Nov 19, 2026
87 of 88 checks passed
@kubestellar-prow
kubestellar-prow Bot deleted the fix/emdash-outbound-durable branch November 19, 2026 01:42
@clubanderson clubanderson added lgtm Indicates that a PR is ready to be merged. approved Indicates a PR has been approved by an approver from all required OWNERS files. labels Nov 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

approved Indicates a PR has been approved by an approver from all required OWNERS files. dco-signoff: yes Indicates the PR's author has signed the DCO. lgtm Indicates that a PR is ready to be merged. size/XXL Denotes a PR that changes 1000+ lines, ignoring generated files.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

🐛 Em-dashes still reach GitHub through agent writes on the egress proxy

1 participant