fix: defense-in-depth against duplicate code agent PRs (#1312, #1320, #1321) by jhutar · Pull Request #2373 · fullsend-ai/fullsend

jhutar · 2026-06-17T07:23:58Z

Summary

Prevents duplicate code agent PRs through three defense layers, all addressing the same 2026-05-21 incident where 5 duplicate PRs were created:

Code agent layer (Code agent skip logic should abort PR creation, not just post a comment #1312): pre-code.sh now writes skip=true to $GITHUB_OUTPUT when an existing PR is found, and reusable-code.yml gates all post-validation steps on steps.validate.outputs.skip != 'true'
Triage agent layer (Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321): Added a hard constraint to the triage agent definition — when an open PR already addresses the issue, use action: "prerequisites" instead of action: "sufficient", preventing the ready-to-code label from being applied
Router/dispatch layer (Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320): Added a pr-check step in both dispatch files (reusable-dispatch.yml and scaffold dispatch.yml) that searches for open PRs mentioning the issue number and skips code dispatch entirely when any are found

Closes #1312
Closes #1320
Closes #1321

Test plan

All 21 pre-code tests pass (18 existing + 3 new covering GITHUB_OUTPUT skip signal)
shellcheck clean on pre-code.sh
actionlint clean on reusable-code.yml and reusable-dispatch.yml
Scaffold workflow alignment tests pass (go test ./internal/scaffold/)
Trigger code agent on an issue with an existing open PR — agent should post skip notice and exit without creating a branch or PR

🤖 Generated with Claude Code

github-actions · 2026-06-17T07:24:10Z

E2E tests are running

Authorization passed for this commit. See the E2E Tests workflow for results.

github-actions · 2026-06-17T07:25:26Z

Site preview

Preview: https://0738e782-site.fullsend-ai.workers.dev

Commit: cf544d0c38f3928817e54edc6d23b064023e22e5

jhutar · 2026-06-17T07:27:13Z

Hello @rh-hemartin . Would it be possible to make me a member of the org so I do not have to wait for /ok-to-test next time? Or if that is not a usual practice, please ignore me.

fullsend-ai-review · 2026-06-17T07:27:24Z

🤖 Finished Review · ✅ Success · Started 7:27 AM UTC · Completed 7:39 AM UTC
Commit: efb551b · View workflow run →

codecov · 2026-06-17T07:31:37Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

fullsend-ai-review · 2026-06-17T07:39:35Z

Review

Findings

Medium

[protected-path] .github/workflows/reusable-code.yml, .github/workflows/reusable-dispatch.yml — These files are under the .github/ protected path. The PR links to issues Code agent skip logic should abort PR creation, not just post a comment #1312, Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320, Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321 and the description explains the rationale (gating downstream steps on validation skip output, adding dispatch-level duplicate-PR check). Human approval is required for protected-path changes regardless of context.
[scope-authorization] .github/workflows/reusable-code.yml — The PR implements three linked issues (Code agent skip logic should abort PR creation, not just post a comment #1312, Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320, Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321) in a single PR spanning six files across three architectural layers (triage agent, dispatch router, code agent workflow). While each change maps to a specific issue, the combined scope creates a large review surface. Consider whether separate PRs would enable cleaner review and rollback.
[architectural-coherence] .github/workflows/reusable-dispatch.yml — The new pr-check step in the dispatch workflow introduces a second location checking for existing PRs, alongside the existing check in pre-code.sh. This is a valid defense-in-depth pattern (failing fast to avoid unnecessary workflow dispatch), but creates maintenance burden — the check logic now lives in two places with slightly different behavior (dispatch uses gh pr list --search, pre-code uses formal PR linking + search with richer labeling/commenting).
[design-coherence] internal/scaffold/fullsend-repo/agents/triage.md — The hard constraint repurposes action: "prerequisites" to mean "an existing PR already addresses this issue." While the prerequisites action's existing array accepts PR URLs and the semantic fit is workable ("wait for this PR to land before starting duplicate work"), it conflates dependency-blocking with duplicate-work-detection. A dedicated action type would be more precise but would require schema and pipeline changes.

Low

[logic-error] .github/workflows/reusable-dispatch.yml:67 — When pr-check sets skipped=true, the stage output is correctly blanked, but intermediate steps (Validate routed stage, Check kill switch, Check role is enabled) still run because their if conditions only check steps.route.outputs.stage != ''. These steps are harmless read-only validations, so the impact is minor inefficiency rather than a correctness bug. The same pattern exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
[edge-case] .github/workflows/reusable-dispatch.yml — The gh pr list --search "${ISSUE_NUMBER} in:title,body" performs a free-text search. For low issue numbers (e.g., docs: Add agent-compatible code problem document #1, Add problem areas: Tekton pipeline review, migration path, multi-tenancy #2), this could match unrelated PRs containing those digits. However, this is a defense-in-depth guard; false positives cause a fail-closed skip (the safe direction), and the primary check in pre-code.sh uses proper author filtering. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
[fail-open] .github/workflows/reusable-dispatch.yml:67 — The pr-check guard fails open: if gh pr list errors out (network failure, rate limit), the || true suppresses the error, MENTIONING_PRS is empty, skipped is never set, and the code stage proceeds. Intentional fail-open for defense-in-depth. The same pattern exists in the scaffold dispatch.yml.
[edge-case] internal/scaffold/fullsend-repo/scripts/pre-code.sh:46 — The validation-failure exit path (exit 1) does not write to $GITHUB_OUTPUT. Not a runtime problem because step failure halts the job by default.
[missing-doc] docs/guides/user/bugfix-workflow.md — The bugfix workflow guide describes the code agent stage but does not mention the prerequisite check that prevents duplicate PRs when an existing human-authored PR is found.

Info

[fail-open] internal/scaffold/fullsend-repo/scripts/pre-code.sh:58 — When GH_TOKEN is unset, the script exits early bypassing the existing-PR check. The PR makes this behavior explicit by writing skipped=false. Intentional best-effort design.
[naming-inconsistency] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — Uses variable name gh_output while pre-fetch-prior-review-test.sh uses github_output for the same purpose.
[missing-doc] docs/agents/code.md — The pr-open label applied by pre-code.sh when an existing human PR is found is not documented in the code agent docs.

Previous run

Review

Findings

Medium

[protected-path] .github/workflows/reusable-code.yml, .github/workflows/reusable-dispatch.yml — These files are under the .github/ protected path. The PR links to issues Code agent skip logic should abort PR creation, not just post a comment #1312, Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320, Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321 and the description explains the rationale (gating downstream steps on validation skip output, adding dispatch-level duplicate-PR check). Human approval is required for protected-path changes regardless of context.

Low

[edge-case] internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml:207 — The gh pr list --search "${ISSUE_NUMBER} in:title,body" performs a free-text search. For low issue numbers (e.g., docs: Add agent-compatible code problem document #1, Add problem areas: Tekton pipeline review, migration path, multi-tenancy #2), this could match unrelated PRs containing those digits. However, this is a defense-in-depth guard; false positives cause a fail-closed skip (the safe direction), and the primary check in pre-code.sh uses proper author filtering. The same issue exists in .github/workflows/reusable-dispatch.yml.
[fail-open] .github/workflows/reusable-dispatch.yml:67 — The pr-check guard fails open: if gh pr list errors out (network failure, rate limit), the || true suppresses the error, MENTIONING_PRS is empty, skipped is never set, and the code stage proceeds. Intentional fail-open for defense-in-depth. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
[missing-doc] docs/guides/user/bugfix-workflow.md — The bugfix workflow guide describes the code agent stage but does not mention the prerequisite check that prevents duplicate PRs when an existing human-authored PR is found.

Info

[edge-case] internal/scaffold/fullsend-repo/scripts/pre-code.sh:46 — The validation-failure exit path (exit 1) does not write to $GITHUB_OUTPUT. Not a runtime problem because step failure halts the job by default.
[fail-open] internal/scaffold/fullsend-repo/scripts/pre-code.sh:58 — When GH_TOKEN is unset, the script exits early bypassing the existing-PR check. The PR makes this behavior explicit by writing skipped=false. Intentional best-effort design.
[naming-inconsistency] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — Uses variable name gh_output while pre-fetch-prior-review-test.sh uses github_output for the same purpose.
[missing-doc] docs/agents/code.md — The pr-open label applied by pre-code.sh when an existing human PR is found is not documented in the code agent docs.

Previous run (2)

Review

Findings

Medium

[logic-error] .github/workflows/reusable-dispatch.yml:247 — The dispatch-level pr-check step does not filter out bot-authored PRs (fullsend-ai[bot], fullsend-ai-coder[bot]), while pre-code.sh explicitly excludes them via --jq select(). If the code agent itself has an open PR for the issue, the dispatch-level check will block a re-run, even though the intent is to only block when a human PR exists. This creates an inconsistency between the two defense layers. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
Remediation: Add --jq filtering to the dispatch-level pr-check step to exclude bot-authored PRs, matching the logic in pre-code.sh.
[protected-path] .github/workflows/reusable-code.yml, .github/workflows/reusable-dispatch.yml — These files are under the .github/ protected path. The PR links to issues Code agent skip logic should abort PR creation, not just post a comment #1312, Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320, Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321 and the description explains the rationale (gating downstream steps on validation skip output and adding dispatch-level duplicate-PR check). Human approval is required for protected-path changes regardless of context.

Low

[edge-case] .github/workflows/reusable-dispatch.yml:247 — The gh pr list --search "${ISSUE_NUMBER} in:title,body" performs a free-text search. For low issue numbers (e.g., docs: Add agent-compatible code problem document #1, Add problem areas: Tekton pipeline review, migration path, multi-tenancy #2), this could match unrelated PRs containing those digits. However, this is a defense-in-depth guard; false positives cause a fail-closed skip (the safe direction), and the primary check in pre-code.sh uses proper author filtering. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
[fail-open] .github/workflows/reusable-dispatch.yml:67 — The pr-check guard fails open: if gh pr list errors out (network failure, rate limit), the || true suppresses the error, MENTIONING_PRS is empty, skipped is never set, and the code stage proceeds. Acceptable for defense-in-depth but worth documenting as intentionally fail-open.

Info

[edge-case] internal/scaffold/fullsend-repo/scripts/pre-code.sh:46 — The validation-failure exit path (exit 1) does not write to $GITHUB_OUTPUT. Not a runtime problem because step failure halts the job by default.
[fail-open] internal/scaffold/fullsend-repo/scripts/pre-code.sh:58 — When GH_TOKEN is unset, the script exits early bypassing the existing-PR check. The PR makes this behavior explicit by writing skipped=false. Intentional best-effort design.
[naming-inconsistency] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — Uses variable name gh_output while pre-fetch-prior-review-test.sh uses github_output for the same purpose.

Previous run (3)

Review

Findings

Medium

[logic-error] .github/workflows/reusable-dispatch.yml:247 — The dispatch-level pr-check step does not filter out bot-authored PRs (fullsend-ai[bot], fullsend-ai-coder[bot]), while pre-code.sh explicitly excludes them via --jq select(). If the code agent itself has an open PR for the issue, the dispatch-level check will block a re-run, even though the intent is to only block when a human PR exists. This creates an inconsistency between the two defense layers. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
Remediation: Add --jq filtering to the dispatch-level pr-check step to exclude bot-authored PRs, matching the logic in pre-code.sh.
[protected-path] .github/workflows/reusable-code.yml, .github/workflows/reusable-dispatch.yml — These files are under the .github/ protected path. The PR links to issues Code agent skip logic should abort PR creation, not just post a comment #1312, Expand ISSUE_HAS_PR check to detect PRs that mention (not just formally link) an issue #1320, Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321 and the description explains the rationale (gating downstream steps on validation skip output and adding dispatch-level duplicate-PR check). Human approval is required for protected-path changes regardless of context.

Low

[edge-case] .github/workflows/reusable-dispatch.yml:247 — The gh pr list --search "${ISSUE_NUMBER} in:title,body" performs a free-text search. For low issue numbers (e.g., docs: Add agent-compatible code problem document #1, Add problem areas: Tekton pipeline review, migration path, multi-tenancy #2), this could match unrelated PRs containing those digits. However, this is a defense-in-depth guard; false positives cause a fail-closed skip (the safe direction), and the primary check in pre-code.sh uses proper author filtering. The same issue exists in internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml.
[fail-open] .github/workflows/reusable-dispatch.yml:67 — The pr-check guard fails open: if gh pr list errors out (network failure, rate limit), the || true suppresses the error, MENTIONING_PRS is empty, skipped is never set, and the code stage proceeds. Acceptable for defense-in-depth but worth documenting as intentionally fail-open.

Info

[edge-case] internal/scaffold/fullsend-repo/scripts/pre-code.sh:46 — The validation-failure exit path (exit 1) does not write to $GITHUB_OUTPUT. Not a runtime problem because step failure halts the job by default.
[fail-open] internal/scaffold/fullsend-repo/scripts/pre-code.sh:58 — When GH_TOKEN is unset, the script exits early bypassing the existing-PR check. The PR makes this behavior explicit by writing skipped=false. Intentional best-effort design.
[naming-inconsistency] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — Uses variable name gh_output while pre-fetch-prior-review-test.sh uses github_output for the same purpose.

Previous run (4)

Review

Findings

Medium

[protected-path] .github/workflows/reusable-code.yml — This file is under the .github/ protected path. The PR links to issue Code agent skip logic should abort PR creation, not just post a comment #1312 and the description explains the rationale (gating downstream steps on the pre-code skip output). Human approval is required for protected-path changes regardless of context.

Low

[edge-case] internal/scaffold/fullsend-repo/scripts/pre-code.sh:46 — The validation-failure exit path (exit 1 on line 46) does not write skip=true to $GITHUB_OUTPUT. While this is not a runtime problem today (step failure halts the job by default), adding echo "skip=true" >> "${GITHUB_OUTPUT}" before exit 1 would make the gate defensive against future continue-on-error additions.
[test-coverage-gap] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — The no-GH_TOKEN exit path writes skip=false to GITHUB_OUTPUT, but the existing test no-gh-token-skips-check only asserts on stdout. Consider adding a run_test_github_output variant for this path.
[naming-inconsistency] .github/workflows/reusable-code.yml — This PR uses output name skip while reusable-dispatch.yml uses skipped for the same conceptual pattern. The inconsistency is cosmetic (different scripts, different step IDs, different semantic contexts) but worth noting for future alignment.

Info

[documentation] internal/scaffold/fullsend-repo/scripts/pre-code.sh — The script header (lines 1–11) documents required environment variables but not the skip output variable it now produces. Consider documenting it, similar to how other pre-scripts document their outputs.
[naming-conventions] internal/scaffold/fullsend-repo/scripts/pre-code-test.sh — Uses variable name gh_output while pre-fetch-prior-review-test.sh uses github_output for the same purpose. Minor naming inconsistency.
[architectural-coherence] internal/scaffold/fullsend-repo/scripts/pre-code.sh — This script now writes GITHUB_OUTPUT on all exit paths, but other pre-* scripts do not. The asymmetry is acceptable since only the code agent currently needs skip signaling.

fullsend-ai-review · 2026-06-17T09:37:54Z

🤖 Review · ⚠️ Cancelled · Started 9:37 AM UTC · Ended 9:43 AM UTC
Commit: 4e21a60 · View workflow run →

fullsend-ai-review · 2026-06-17T09:47:17Z

🤖 Review · ❌ Terminated · Started 9:47 AM UTC · Ended 9:59 AM UTC
Commit: 4e21a60 · View workflow run →

jhutar · 2026-06-17T09:47:55Z

Addressed 2 review comments with 2 additional PRs.

fullsend-ai-review

See the review comment for full details.

fullsend-ai-review · 2026-06-17T09:59:03Z

🤖 Finished Review · ✅ Success · Started 9:47 AM UTC · Completed 9:59 AM UTC
Commit: 481de7c · View workflow run →

jhutar · 2026-06-17T10:26:25Z

Added another commit to fix another review comment.

fullsend-ai-review · 2026-06-17T10:29:34Z

🤖 Finished Review · ✅ Success · Started 10:29 AM UTC · Completed 10:42 AM UTC
Commit: 2c94f53 · View workflow run →

fullsend-ai-review

See the review comment for full details.

rh-hemartin · 2026-06-18T10:19:51Z

You need to signoff all the commits.

pre-code.sh correctly detected existing PRs and posted a skip comment, but exited 0 without signaling the workflow to stop — so all downstream steps (GCP setup, bot identity, agent run) executed anyway, producing duplicate PRs. Write skip=true/false to GITHUB_OUTPUT on every exit path and gate all post-validation steps on steps.validate.outputs.skip != 'true'. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

The triage agent correctly identified existing PRs during its search but still emitted action "sufficient", applying ready-to-code and triggering duplicate code agent dispatches. Add a hard constraint in Step 2b: when an open PR already addresses the issue, use action "prerequisites" with the PR URL instead of "sufficient". Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

…ssue The dispatch router had no check for existing PRs that reference an issue without formal closing keywords. Add a pr-check step in both dispatch files (reusable-dispatch.yml and scaffold dispatch.yml) that searches for open PRs mentioning the issue number and skips code dispatch when any are found. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

…tests The no-token exit path writes skip=false to GITHUB_OUTPUT but the existing test only asserted on stdout. Add a run_test_github_output variant to verify the output file. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

Align with the existing convention used by role-check steps in the dispatch workflows, which output skipped=true. Rename skip→skipped in pre-code.sh, reusable-code.yml, reusable-dispatch.yml, scaffold dispatch.yml, and corresponding tests. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

…r-check The dispatch pr-check step did not filter out fullsend-ai[bot] and fullsend-ai-coder[bot] PRs, which would block re-runs even when only a bot PR existed — making the /fs-code --force escape hatch unreachable. Add --jq filtering to match the logic in pre-code.sh. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Generated-by: Claude rh-pre-commit.version: 2.4.0 rh-pre-commit.check-secrets: ENABLED Signed-off-by: Jan Hutar <jhutar@redhat.com>

jhutar · 2026-06-18T11:47:25Z

Oh, sorry, missed it. All commits signed now.

fullsend-ai-review · 2026-06-18T11:50:27Z

🤖 Finished Review · ✅ Success · Started 11:50 AM UTC · Completed 12:04 PM UTC
Commit: cf544d0 · View workflow run →

fullsend-ai-review

See the review comment for full details.

fullsend-ai-review · 2026-06-18T12:04:25Z

      pull-requests: read
    outputs:
-      stage: ${{ steps.role-check.outputs.skipped != 'true' && steps.route.outputs.stage || '' }}
+      stage: ${{ steps.role-check.outputs.skipped != 'true' && steps.pr-check.outputs.skipped != 'true' && steps.route.outputs.stage || '' }}


[low] logic-error

When pr-check sets skipped=true, intermediate steps still run. Harmless read-only validations with minor inefficiency.

fullsend-ai-review · 2026-06-18T12:04:25Z

      pull-requests: read
    outputs:
-      stage: ${{ steps.role-check.outputs.skipped != 'true' && steps.route.outputs.stage || '' }}
+      stage: ${{ steps.role-check.outputs.skipped != 'true' && steps.pr-check.outputs.skipped != 'true' && steps.route.outputs.stage || '' }}


[low] fail-open

The pr-check guard fails open on API errors. Intentional fail-open for defense-in-depth.

fullsend-ai-retro · 2026-06-18T13:33:04Z

🤖 Finished Retro · ✅ Success · Started 1:33 PM UTC · Completed 1:45 PM UTC
Commit: cf544d0 · View workflow run →

fullsend-ai-retro · 2026-06-18T13:45:48Z

Retro: PR #2373 — defense-in-depth against duplicate code agent PRs

Timeline

Human-authored PR by jhutar, opened 2026-06-17, merged 2026-06-18 (~30 hours). The PR implemented three defense layers against duplicate code agent PRs (closing #1312, #1320, #1321). Six commits across 6 files (+149/-8). Two human approvals (rh-hemartin, ralphbean).

Agent activity: 7 review dispatches → 4 successful, 2 cancelled (rapid pushes), 1 cancelled (superseded). Five distinct review iterations produced findings across the PR's evolution.

What went well

Review caught a real bug. The medium-severity [logic-error] finding on reusable-dispatch.yml — that the dispatch-level pr-check didn't filter out bot-authored PRs while pre-code.sh did — was a genuine correctness issue. The author fixed it in commit cf544d0.
Author was responsive. Addressed the review finding promptly, responded to lower-severity comments with reasoning, and signed off commits when asked.
Reasonable time to merge for a multi-issue fix touching workflow files.

Known inefficiencies (all tracked by existing issues)

Repeated findings across iterations: Many low/info findings (fail-open, edge-case, naming-inconsistency) persisted across all 5 review runs without new signal. Already tracked: Review agent should deduplicate findings across iterations on the same PR #1013, Review agent should not regenerate unchanged inline comments on re-reviews #1285, Review agent: resolve inline comments from previous reviews on re-review #956, Review agent should not re-request changes for unchanged findings #1500, Review agent should carry forward previous findings when re-reviewing unchanged diffs #1979.
Duplicate review run on same commit: Runs on 481de7c produced byte-identical findings twice. Already tracked: Deduplicate review dispatches for the same HEAD SHA #1452, Deduplicate review runs when HEAD SHA has not changed #1567.
New findings in final run after human approval: Three new medium-severity findings (scope-authorization, architectural-coherence, design-coherence) appeared in the last review iteration after both humans had already approved. Already tracked: Review agent surfaces new low-severity findings across multiple approved passes #1367, Review bot should not override human approval with stale findings #1922.
Cancelled runs from rapid pushes: 2 of 7 dispatches were cancelled. Already tracked: Enable cancel-in-progress for review dispatches on the same PR #1357, Debounce review dispatch on rapid synchronize events #1014, Deduplicate review runs on rapid successive pushes #1418.
Info-severity inline comments: Review posted inline comments for info-level findings, adding noise. Already tracked: Review agent: never post inline comments for info-severity findings #2287.

Conclusion

The workflow delivered good value — the review agent found and the author fixed a real bug. The inefficiencies observed (repeated findings, duplicate runs, late-appearing findings) are well-known patterns with extensive existing issue coverage. No new proposals are warranted.

The dispatch "Check for existing PRs" guard used --search "N in:title,body", which matches any PR that mentions the issue number anywhere in its body — including unrelated cross-references — and skipped code dispatch whenever one existed. Its bot-authored-PR exclusion also never worked: GraphQL's Bot.login omits the REST "[bot]" suffix, so the literal string comparison against "fullsend-ai-coder[bot]" never matched. Replace both with a GraphQL query for closedByPullRequestsReferences, which only returns PRs that actually close the issue via Fixes/Closes/ Resolves keywords, and match the bot exclusion on __typename == "Bot" instead of a REST-formatted login string. The query also returns MERGED PRs regardless of includeClosedPrs, so filter on .state == "OPEN" explicitly — otherwise a long-merged closer would permanently block re-dispatch on a reopened issue. Add issues: read to both jobs' permissions, since the query now resolves through the Issue type rather than pull-requests-only fields. Separate stderr from the query's stdout so an incidental warning on an otherwise-successful call can't pollute the result and trigger a false-positive skip. internal/scaffold/fullsend-repo/scripts/pre-code.sh had the identical bug (same substring search, same broken bot-login comparison) and runs as a second gate in the same code-dispatch pipeline, so the original false-positive-skip failure mode could still recur through that path. Apply the same fix there and correct pre-code-test.sh's mock fixtures, which hardcoded the REST-suffixed bot login format and so validated the wrong assumption about what the API actually returns. Verified against production data: #5569 (false positive: PR #5192 only mentions the issue, now correctly proceeds), #1320 (historical: closed by now-merged PR #2373, now correctly proceeds since the closer is no longer open), #5560 (bot-authored closer, correctly excluded), and #5575 itself (open, human-authored PR #5578 with "Fixes #5575" in its body, correctly detected as blocking). Assisted-by: Claude Signed-off-by: Wayne Sun <gsun@redhat.com>

The dispatch "Check for existing PRs" guard used --search "N in:title,body", which matches any PR that mentions the issue number anywhere in its body — including unrelated cross-references — and skipped code dispatch whenever one existed. Its bot-authored-PR exclusion also never worked: GraphQL's Bot.login omits the REST "[bot]" suffix, so the literal string comparison against "fullsend-ai-coder[bot]" never matched. Replace both with a GraphQL query for closedByPullRequestsReferences, which only returns PRs that actually close the issue via Fixes/Closes/ Resolves keywords, and match the bot exclusion on __typename == "Bot" instead of a REST-formatted login string. The query also returns MERGED PRs regardless of includeClosedPrs, so filter on .state == "OPEN" explicitly — otherwise a long-merged closer would permanently block re-dispatch on a reopened issue. Add issues: read to both jobs' permissions, since the query now resolves through the Issue type rather than pull-requests-only fields. Separate stderr from the query's stdout so an incidental warning on an otherwise-successful call can't pollute the result and trigger a false-positive skip. internal/scaffold/fullsend-repo/scripts/pre-code.sh had the identical bug (same substring search, same broken bot-login comparison) and runs as a second gate in the same code-dispatch pipeline, so the original false-positive-skip failure mode could still recur through that path. Apply the same fix there and correct pre-code-test.sh's mock fixtures, which hardcoded the REST-suffixed bot login format and so validated the wrong assumption about what the API actually returns. Verified against production data: #5569 (false positive: PR #5192 only mentions the issue, now correctly proceeds), #1320 (historical: closed by now-merged PR #2373, correctly proceeds since the closer is no longer open), #5560 (bot-authored closer, correctly excluded), and #5575 itself (open, human-authored PR #5578 with a "Fixes" keyword, correctly detected as blocking). Also bump closedByPullRequestsReferences's first from 20 to 100 (the connection's API max) at all three call sites, since a long-lived, repeatedly-reopened issue could otherwise silently truncate past the 20th closing-PR reference; add null-safety around the nodes array and author login so a missing field degrades gracefully instead of erroring or printing "null"; sanitize captured stderr before interpolating it into a ::warning:: workflow command; fix a pre-code-test.sh case that mocked a query failure while asserting the "no linked PRs" behavior; and document a third bot-login format returned by gh's own --json output (app/<slug> with a separate is_bot flag). Assisted-by: Claude (fix, review), Grok (review) Signed-off-by: Wayne Sun <gsun@redhat.com>

github-actions Bot deployed to site-preview June 17, 2026 07:25 View deployment

jhutar mentioned this pull request Jun 17, 2026

Triage agent should not label ready-to-code when it identifies an existing PR that addresses the issue #1321

Closed

fullsend-ai-review Bot added the requires-manual-review Review requires human judgment label Jun 17, 2026

rh-hemartin assigned jhutar Jun 17, 2026

rh-hemartin added the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Jun 17, 2026

github-actions Bot removed the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Jun 17, 2026

github-actions Bot deployed to site-preview June 17, 2026 09:35 View deployment

jhutar changed the title ~~fix(#1312): gate code agent steps on pre-code skip output~~ fix: defense-in-depth against duplicate code agent PRs (#1312, #1320, #1321) Jun 17, 2026

jhutar force-pushed the 3-no-code-on-skip branch from 514a886 to 481de7c Compare June 17, 2026 09:43

github-actions Bot deployed to site-preview June 17, 2026 09:45 View deployment

fullsend-ai-review Bot reviewed Jun 17, 2026

View reviewed changes

Comment thread .github/workflows/reusable-dispatch.yml

Comment thread .github/workflows/reusable-dispatch.yml

Comment thread .github/workflows/reusable-dispatch.yml

Comment thread internal/scaffold/fullsend-repo/scripts/pre-code.sh

fullsend-ai-review Bot added requires-manual-review Review requires human judgment and removed requires-manual-review Review requires human judgment labels Jun 17, 2026

github-actions Bot deployed to site-preview June 17, 2026 10:27 View deployment

fullsend-ai-review Bot reviewed Jun 17, 2026

View reviewed changes

Comment thread internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml

Comment thread .github/workflows/reusable-dispatch.yml

Comment thread internal/scaffold/fullsend-repo/scripts/pre-code.sh

fullsend-ai-review Bot added requires-manual-review Review requires human judgment and removed requires-manual-review Review requires human judgment labels Jun 17, 2026

guyoron1 mentioned this pull request Jun 17, 2026

fix: defense-in-depth against duplicate code agent PRs (#1312, #1320, #1321) guyoron1/fullsend#26

Closed

rh-hemartin added the ok-to-test Allow e2e CI to run after maintainer review (must be re-applied after each push) label Jun 18, 2026

rh-hemartin approved these changes Jun 18, 2026

View reviewed changes

jhutar added 6 commits June 18, 2026 13:46

jhutar force-pushed the 3-no-code-on-skip branch from 2c94f53 to cf544d0 Compare June 18, 2026 11:46

github-actions Bot deployed to site-preview June 18, 2026 11:48 View deployment

fullsend-ai-review Bot reviewed Jun 18, 2026

View reviewed changes

fullsend-ai-review Bot added requires-manual-review Review requires human judgment and removed requires-manual-review Review requires human judgment labels Jun 18, 2026

ralphbean approved these changes Jun 18, 2026

View reviewed changes

ralphbean added this pull request to the merge queue Jun 18, 2026

Merged via the queue into fullsend-ai:main with commit 24f969e Jun 18, 2026
16 checks passed

This was referenced Jul 14, 2026

fix(#4718): stop running pre-code/pre-fix scripts twice per run #4762

Closed

fix(#4718): wire GH_TOKEN and re-add fail-fast validation for pre-code/pre-fix #5013

Closed

waynesun09 mentioned this pull request Jul 24, 2026

fix(#5575): match linked-PR closing keywords instead of substring search #5578

Merged

Uh oh!

Conversation

jhutar commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

github-actions Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

E2E tests are running

Uh oh!

github-actions Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Site preview

Uh oh!

jhutar commented Jun 17, 2026

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Jun 17, 2026

Codecov Report

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review

Findings

Medium

Low

Info

Review

Findings

Medium

Low

Info

Review

Findings

Medium

Low

Info

Review

Findings

Medium

Low

Info

Review

Findings

Medium

Low

Info

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jhutar commented Jun 17, 2026

Uh oh!

fullsend-ai-review Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026

Uh oh!

jhutar commented Jun 17, 2026

Uh oh!

fullsend-ai-review Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fullsend-ai-review Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

rh-hemartin commented Jun 18, 2026

jhutar commented Jun 17, 2026 •

edited

Loading

github-actions Bot commented Jun 17, 2026 •

edited

Loading

github-actions Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 17, 2026 •

edited

Loading

fullsend-ai-review Bot commented Jun 18, 2026 •

edited

Loading

fullsend-ai-retro Bot commented Jun 18, 2026 •

edited

Loading