chore: update tend workflows (0.1.14 → 0.1.15) (#3814)

Automated nightly regeneration of tend's workflow files. **tend version:** 0.1.14 → 0.1.15 **Notable changes** - **Rate-limit handling**: a maintainer can approve past the spike limit (max-sixty/tend#874), and `tend-mention` now skips comments on `tend-rate-limit`-labelled issues as well as `tend-outage` ones — the same self-trigger loop guard, widened to every issue tend files about its own health. - **Review flow**: a push is queued behind an examined-HEAD gate instead of cancelling the in-flight review (max-sixty/tend#903), the review is submitted before a fix is pushed (max-sixty/tend#834), a force-push triggers a re-review rather than trusting the re-anchored SHA (max-sixty/tend#884), and the `/code-review` second pass is unconditional (max-sixty/tend#937). - **CI monitoring**: both the check poll and the `gh run rerun --failed` poll now end terminally when the cap is hit instead of reading as done (max-sixty/tend#876, max-sixty/tend#951), and a failed GitHub-status probe no longer reads as "no incident" (max-sixty/tend#913). - **Nightly**: conflicted bot PRs are test-merged locally instead of filtered on the lazy `mergeable` field (max-sixty/tend#898), and mention runs skip the bot's own review and a third party's content-free approval (max-sixty/tend#916, max-sixty/tend#955). - **`tend check`**: an unreadable ruleset bypass list is reported as unknown rather than ungated (max-sixty/tend#825), environment names are read one per line and addressed encoded (max-sixty/tend#879), and `credential-environments` no longer points at the setting it rejects (max-sixty/tend#900) — this repo currently `SKIP`s that check, so its wording should improve here. Full comparison: https://github.com/max-sixty/tend/compare/0.1.14...0.1.15 Co-authored-by: worktrunk-bot <254187624+worktrunk-bot@users.noreply.github.com>
2026-09-14 20:00:38 +08:00 · 2026-08-13 05:41:10 -07:00
parent 7aba380f0c
commit c31e827cba
8 changed files with 201 additions and 35 deletions
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -34,7 +34,7 @@ jobs:

      - uses: ./.github/actions/claude-setup

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -90,20 +90,20 @@ jobs:
            -f "client_payload[id]=${{ github.event.review.id || github.event.comment.id }}"

  verify:
-    # Skip comments on `tend-outage` issues: the action's Report-failure step
-    # auto-comments on those when Claude invocation fails, and without this
-    # guard those comments re-trigger tend-mention during a persistent outage
-    # (e.g. Anthropic 401), producing a self-sustaining ~1 run/minute loop
-    # until the outage clears. The prompt's self-loop guard can't help here
-    # because the model never executes — the action fails before Claude starts.
-    # A relayed review enters as `repository_dispatch` and is judged in the
-    # check step below, against the record the API holds.
+    # Skip comments on the issues tend files about its own health: the action
+    # auto-comments on those when a run fails or is refused, and without this
+    # guard those comments re-trigger tend-mention, producing a
+    # self-sustaining ~1 run/minute loop until the underlying condition
+    # clears. The prompt's self-loop guard can't help here because the model
+    # never executes — the action fails before Claude starts. A relayed review
+    # enters as `repository_dispatch` and is judged in the check step below,
+    # against the record the API holds.
    if: |
      github.event_name == 'repository_dispatch' ||
      (github.event_name == 'issues' &&
        contains(github.event.issue.body, '@worktrunk-bot')) ||
      (github.event_name == 'issue_comment' &&
-        !contains(github.event.issue.labels.*.name, 'tend-outage'))
+        contains(github.event.issue.labels.*.name, 'tend-outage') == false && contains(github.event.issue.labels.*.name, 'tend-rate-limit') == false)
    runs-on: ubuntu-24.04
    environment:
      name: tend
@@ -195,8 +195,11 @@ jobs:
          # comments on purpose: the pull_request_review *submission* kind is
          # deliberately left out, since a review the bot leaves on its own PR
          # is its reviewer role (the prompt is told to action it), not a
-          # self-loop. The only self-review skip is the terminal empty-body
-          # APPROVED gate below.
+          # self-loop. The three self-review skips below are author-keyed, but
+          # each is also narrowed to a case that leaves this run nothing to do:
+          # the terminal empty-body APPROVED gate, the synthetic reply
+          # container, and a bot review on a PR the bot did not author. None of
+          # them licenses a blanket self-review skip.
          if { [ "$KIND" = "issue_comment" ] || [ "$KIND" = "pull_request_review_comment" ]; } \
             && [ "$COMMENT_AUTHOR" = "worktrunk-bot" ]; then
            echo "should_run=false" >> "$GITHUB_OUTPUT"
@@ -242,14 +245,58 @@ jobs:
          # A review's record includes review.body (checked above) but NOT the
          # bodies of the inline comments attached to the review. Fetch them
          # so a first-contact @-mention inside an inline comment is detected
-          # on PRs where the bot has no prior engagement.
+          # on PRs where the bot has no prior engagement. One object per line,
+          # so `--paginate` concatenates pages instead of reducing within one.
+          # Keep the `{body, in_reply_to_id}` construction: `in_reply_to_id` is
+          # an *optional* property, absent rather than null on a fresh comment,
+          # and building the object normalizes absent to null so the `== null`
+          # select below counts both shapes. A bare `.in_reply_to_id` stream
+          # would emit nothing for a fresh comment and skip every container.
          if [ "$KIND" = "pull_request_review" ]; then
-            if gh api --paginate "repos/$GITHUB_REPOSITORY/pulls/$PAYLOAD_PR/reviews/$PAYLOAD_ID/comments" \
-                 --jq '.[].body' | grep -qF '@worktrunk-bot'; then
+            INLINE=$(gh api --paginate "repos/$GITHUB_REPOSITORY/pulls/$PAYLOAD_PR/reviews/$PAYLOAD_ID/comments" \
+              --jq '.[] | {body, in_reply_to_id}')
+
+            if printf '%s\n' "$INLINE" | jq -r '.body' | grep -qF '@worktrunk-bot'; then
              echo "should_run=true" >> "$GITHUB_OUTPUT"
              echo "reason=mention" >> "$GITHUB_OUTPUT"
              exit 0
            fi
+
+            # Replying to a review thread wraps the reply in a synthetic
+            # zero-body COMMENTED review, so the bot's own inline reply arrives
+            # here as a `pull_request_review` submission as well as the
+            # `pull_request_review_comment` the skip above already drops. Same
+            # artifact, second event path: without this the engagement heuristic
+            # below sees a bot-authored PR (or BOT_REVIEWS > 0) and spins up a
+            # full handle job that the prompt's self-loop guard then exits.
+            # Narrower than the comment skip on purpose — a real review the bot
+            # submits with inline comments and no body carries actionable signal,
+            # so require that *every* inline comment be a reply
+            # (`in_reply_to_id` set). A container with any fresh inline comment
+            # still fires.
+            if [ "$REVIEW_AUTHOR" = "worktrunk-bot" ] \
+               && [ -z "$COMMENT_BODY" ] \
+               && [ "$(printf '%s\n' "$INLINE" | jq -s '[.[] | select(.in_reply_to_id == null)] | length')" = "0" ]; then
+              echo "should_run=false" >> "$GITHUB_OUTPUT"
+              exit 0
+            fi
+
+            # Somebody else's content-free approval is terminal too. The gate
+            # above is author-keyed, so a human's bare APPROVED falls through to
+            # the PR-author short-circuit (on a bot-authored PR) or to
+            # BOT_REVIEWS (on one the bot has reviewed) and starts a session
+            # whose only possible outcome is a silent exit: an approval with no
+            # body and no inline comments asks for nothing, and the bot cannot
+            # merge on its own. Unlike the bot-authored gates this one requires
+            # `$INLINE` to be empty rather than reply-only — an approval whose
+            # nits live inline is a request to the PR's author, which on a
+            # bot-authored PR is a role the bot has to act in.
+            if [ "$REVIEW_STATE" = "approved" ] \
+               && [ -z "$COMMENT_BODY" ] \
+               && [ -z "$INLINE" ]; then
+              echo "should_run=false" >> "$GITHUB_OUTPUT"
+              exit 0
+            fi
          fi

          # Non-mention: check bot engagement
@@ -290,6 +337,26 @@ jobs:
            echo "reason=participation" >> "$GITHUB_OUTPUT"; exit 0
          fi

+          # A review the bot leaves on someone else's PR leaves this run
+          # nothing to do: whatever the review warranted, the tend-review
+          # session that submitted it has already done — left the findings for
+          # a human author to act on (pushing to their branch unbidden is
+          # barred by conduct rules), or, on a dependency-bot PR where no
+          # author will act, pushed the fix itself. The BOT_REVIEWS heuristic
+          # below counts this very review, so without this gate the session
+          # always starts and always exits silently. Keyed on author alone —
+          # the APPROVED + empty-body gate above is the same shape narrowed to
+          # its one terminal leg, and reusing those clauses here would let every
+          # bodied COMMENTED review through. Placement carries the rest of the
+          # design: *after* the PR_AUTHOR short-circuit, which has already
+          # exited when the PR is the bot's own, so the reviewer-to-author
+          # handoff on a bot PR still fires; *after* the body and inline
+          # @-mention scans, so an explicit summons still wins.
+          if [ "$KIND" = "pull_request_review" ] \
+             && [ "$REVIEW_AUTHOR" = "worktrunk-bot" ]; then
+            echo "should_run=false" >> "$GITHUB_OUTPUT"; exit 0
+          fi
+
          # Captured, not counted — see the note on the issue-comment lookup above.
          BOT_REVIEWS=$(gh api --paginate "repos/$GITHUB_REPOSITORY/pulls/$PR_NUMBER/reviews" \
            --jq '.[] | select(.user.login == "worktrunk-bot") | .id')
@@ -391,7 +458,7 @@ jobs:
          # the API record — the dispatch payload never carries one to spoof.
          EVENT_TS: ${{ github.event.comment.updated_at || needs.verify.outputs.ts || github.event.issue.updated_at }}

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -420,6 +487,13 @@ jobs:
              || format('A user commented on an issue/PR where you previously participated ({0}). Read the full context. Only respond if the comment is directed at you, asks a question you can help with, or requests changes you can make. If the conversation is between other participants, exit silently.', github.event.comment.html_url)
            }}

+      - name: Restore local setup actions for POST cleanup
+        if: always()
+        run: |
+          dir=.github/actions/claude-setup
+          git checkout "$GITHUB_SHA" -- "$dir" ||
+            echo "::warning::could not restore $dir from $GITHUB_SHA; POST cleanup of the local action may fail"
+
      - name: Remove eyes reaction
        if: |
          always()
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -34,7 +34,7 @@ jobs:

      - uses: ./.github/actions/claude-setup

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -27,7 +27,19 @@ jobs:
    steps:
      - name: Check for unread notifications
        id: check
+        env:
+          GITHUB_TOKEN: ${{ secrets.TEND_BOT_TOKEN }}
+          BOT_NAME: worktrunk-bot
        run: |
+          # shellcheck shell=bash
+          # Pre-check for tend-notifications: decide whether the agent needs to boot at
+          # all, and clear inbox noise no agent run is needed for.
+          #
+          # Inlined into the generated workflow (adopter repos have no copy of this
+          # file), so it stays self-contained: env in, GITHUB_OUTPUT out.
+          #
+          # env: BOT_NAME, GITHUB_REPOSITORY, GITHUB_OUTPUT, GITHUB_TOKEN
+
          # Fetch notifications once, tolerating transient non-JSON responses.
          # GitHub occasionally returns an HTML error page (even with a 200)
          # during a brief API blip; under `bash -e` an untolerated `gh api`
@@ -95,7 +107,7 @@ jobs:
            PR_INFO=$(gh api "repos/$GITHUB_REPOSITORY/pulls/$PR_NUM" --jq '"\(.user.login) \(.state)"' 2>/dev/null) || continue
            PR_AUTHOR=${PR_INFO%% *}
            PR_STATE=${PR_INFO##* }
-            if [ "$PR_AUTHOR" = "worktrunk-bot" ] && [ "$PR_STATE" = "closed" ]; then
+            if [ "$PR_AUTHOR" = "$BOT_NAME" ] && [ "$PR_STATE" = "closed" ]; then
              gh api "notifications/threads/$tid" -X PATCH || true
            fi
          done
@@ -125,8 +137,6 @@ jobs:
          else
            echo "$COUNT processable notification(s) — proceeding"
          fi
-        env:
-          GITHUB_TOKEN: ${{ secrets.TEND_BOT_TOKEN }}

      - uses: actions/checkout@v7
        if: steps.check.outputs.count != '0' || github.event_name == 'workflow_dispatch'
@@ -138,7 +148,7 @@ jobs:

      - uses: ./.github/actions/claude-setup
        if: steps.check.outputs.count != '0' || github.event_name == 'workflow_dispatch'
-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        if: steps.check.outputs.count != '0' || github.event_name == 'workflow_dispatch'
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -34,7 +34,7 @@ jobs:

      - uses: ./.github/actions/claude-setup

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -15,7 +15,11 @@ jobs:
  review:
    concurrency:
      group: ${{ github.workflow }}-${{ github.event.pull_request.number }}
-      cancel-in-progress: true
+      # A push mid-review queues a replacement run rather than killing the
+      # session: the running review folds the push in and stamps the commits
+      # it examined, and the gate step below lets the queued run exit without
+      # booting an agent when its HEAD is already covered.
+      cancel-in-progress: false
    runs-on: ubuntu-24.04
    environment:
      name: tend
@@ -26,16 +30,84 @@ jobs:
      actions: read
      issues: write
    steps:
+      - name: Skip when the live HEAD is already examined
+        id: gate
+        env:
+          GITHUB_TOKEN: ${{ secrets.TEND_BOT_TOKEN }}
+          PR: ${{ github.event.pull_request.number }}
+          EVENT_ACTION: ${{ github.event.action }}
+        run: |
+          # shellcheck shell=bash
+          # Pre-check for tend-review: decide whether the agent needs to boot at all.
+          #
+          # The review job runs without cancel-in-progress, so a push mid-review queues
+          # a replacement run while the live session keeps going, folds the push in, and
+          # stamps each commit it examined with a `tend-review/<pr>` commit status (review
+          # skill, "Stamp examined HEADs"). The concurrency group holds this run until
+          # that session ends; by then the live HEAD is usually stamped and there is
+          # nothing left to do. Judged against the live PR, not the event payload — a
+          # queued run's payload is stale by construction.
+          #
+          # Inlined into the generated workflow (adopter repos have no copy of this
+          # file), so it stays self-contained: env in, GITHUB_OUTPUT out. Any write-scoped
+          # actor could forge the stamp to suppress a review; that actor can already
+          # cancel the run itself, so the merge gate — not this check — remains the
+          # security boundary.
+          #
+          # env: PR, EVENT_ACTION, GITHUB_REPOSITORY, GITHUB_OUTPUT, GITHUB_TOKEN
+
+          # Only `synchronize` can be a stale duplicate of an examination that already
+          # happened: `opened` has no prior run, and `reopened` / `ready_for_review`
+          # ask for a fresh pass even on a stamped commit.
+          if [ "$EVENT_ACTION" != "synchronize" ]; then
+            echo "should_run=true" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          # Fail open on API errors: a redundant agent run beats a silently skipped
+          # review. The parse belongs inside the guard — GitHub sometimes returns an
+          # HTML error page with a 200 during a blip, so a zero `gh` exit doesn't mean
+          # the body is JSON, and an unguarded `jq` under the run block's `bash -e`
+          # would fail the step (fail-closed) instead.
+          if ! PR_INFO=$(gh api "repos/$GITHUB_REPOSITORY/pulls/$PR" 2>/dev/null) \
+            || ! STATE=$(echo "$PR_INFO" | jq -re '.state'); then
+            echo "PR #$PR fetch failed — proceeding without the pre-check"
+            echo "should_run=true" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+          if [ "$STATE" != "open" ]; then
+            echo "PR #$PR is $STATE — skipping"
+            echo "should_run=false" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          # The stamp context carries the PR number: one branch can be two open PRs
+          # (same head, different base), and each base means a different diff, so an
+          # examination of one must not gate the other.
+          HEAD=$(echo "$PR_INFO" | jq -r '.head.sha')
+          STAMPED=$(gh api "repos/$GITHUB_REPOSITORY/commits/$HEAD/status?per_page=100" 2>/dev/null \
+            | jq --arg ctx "tend-review/$PR" \
+              '[.statuses[]? | select(.context == $ctx and .state == "success")] | length' \
+            || echo 0)
+          if [ "${STAMPED:-0}" -gt 0 ]; then
+            echo "HEAD $HEAD already examined (tend-review/$PR) — skipping"
+            echo "should_run=false" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          echo "should_run=true" >> "$GITHUB_OUTPUT"
      # Two checkouts: `setup:` runs against the base tree, and the PR's own
      # tree lands after it. Setup executes as the runner user, outside the
      # containment the harness builds for the contributor's code.
      - uses: actions/checkout@v7
+        if: steps.gate.outputs.should_run == 'true'
        with:
          fetch-depth: 0
          fetch-tags: true
          token: ${{ secrets.TEND_BOT_TOKEN }}

      - uses: ./.github/actions/claude-setup
+        if: steps.gate.outputs.should_run == 'true'

      # GitHub only materializes refs/pull/N/merge for mergeable PRs — on
      # conflicting PRs it 404s and every downstream step cascades as skipped.
@@ -44,6 +116,7 @@ jobs:
      # tree.
      - name: Resolve PR checkout ref
        id: pr_ref
+        if: steps.gate.outputs.should_run == 'true'
        env:
          GITHUB_TOKEN: ${{ secrets.TEND_BOT_TOKEN }}
          PR: ${{ github.event.pull_request.number }}
@@ -55,6 +128,7 @@ jobs:
            echo "::notice::refs/pull/$PR/merge unavailable (likely merge conflict); falling back to /head"
          fi
      - uses: actions/checkout@v7
+        if: steps.gate.outputs.should_run == 'true'
        with:
          ref: ${{ steps.pr_ref.outputs.ref }}
          allow-unsafe-pr-checkout: true
@@ -63,7 +137,8 @@ jobs:
          fetch-tags: true
          token: ${{ secrets.TEND_BOT_TOKEN }}

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
+        if: steps.gate.outputs.should_run == 'true'
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -72,3 +147,10 @@ jobs:
          model: opus
          prompt: >-
            ${{ format('/tend-ci-runner:review {0}', github.event.pull_request.number) }}
+
+      - name: Restore local setup actions for POST cleanup
+        if: always() && (steps.gate.outputs.should_run == 'true')
+        run: |
+          dir=.github/actions/claude-setup
+          git checkout "$GITHUB_SHA" -- "$dir" ||
+            echo "::warning::could not restore $dir from $GITHUB_SHA; POST cleanup of the local action may fail"
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -17,7 +17,7 @@ concurrency:

 jobs:
  triage:
-    if: github.repository_owner == 'max-sixty' && contains(github.event.issue.labels.*.name, 'tend-outage') == false
+    if: github.repository_owner == 'max-sixty' && contains(github.event.issue.labels.*.name, 'tend-outage') == false && contains(github.event.issue.labels.*.name, 'tend-rate-limit') == false
    runs-on: ubuntu-24.04
    environment:
      name: tend
@@ -37,7 +37,7 @@ jobs:

      - uses: ./.github/actions/claude-setup

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
@@ -1,4 +1,4 @@
-# Generated by tend 0.1.14. Regenerate with: uvx tend@latest init
+# Generated by tend 0.1.15. Regenerate with: uvx tend@latest init
 #
 # Do not edit this file directly — it will be overwritten on regeneration.
 # To customize behavior, edit the relevant skill (for example,
@@ -34,7 +34,7 @@ jobs:

      - uses: ./.github/actions/claude-setup

-      - uses: max-sixty/tend/claude@0.1.14
+      - uses: max-sixty/tend/claude@0.1.15
        with:
          github_token: ${{ secrets.TEND_BOT_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}