You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The "policy-driven decline reported as hard failure" bug in safe-outputs (push_to_pull_request_branch and approve_workflow_run) has been proposed-but-unshipped since 2026-08-28. Today it hit a real production PR for the first time: run 33921923497 on PR #58608 (Design Decision Gate) failed — including on retry (run_attempt 2) — because the changeset touched files outside the workflow's allowed-files list. That decline is correctly detected but surfaced through the same hard-failure path (auto-filed as #58619, expiring today) as a genuine error, instead of a distinct "declined by policy" outcome.
This is the 6th occurrence of the pattern for Design Decision Gate alone (prior 5 were on internal/tooling PRs only), per today's Safe Output Health Monitor (discussion #58731). The original tracking issue (#56576, proposed 2026-08-28) auto-expired not_planned without a code change; a related tracker (#57850) is no longer resolvable (404). Re-filing to restore a live tracker, now with confirmed production impact.
Expected Impact
Prevents legitimate PRs from being blocked/retried by a false-failure signal, and stops this recurring pattern from polluting safe-output failure-rate metrics across every workflow type (also affects approve_workflow_run's permission-scope failures, which share the same root classification gap).
Suggested Fix
Reclassify policy-driven declines (allowed-files, protected-files) as a distinct non-failure/soft-skip outcome across all safe-output types, not just where partially handled today — likely in the shared safe-outputs processing path (pkg/workflow safe_outputs job / push_to_pull_request_branch handler).
Suggested Agent
Copilot SWE Agent or any coding agent with pkg/workflow write access.
Estimated Effort
Medium (1-4 hours) — touches shared safe-output result classification, needs care to not swallow genuine failures.
Data Source
DeepReport analysis 2026-09-05, sourced from Safe Output Health Monitor discussion #58731 (Cluster 1) and Design Decision Gate failure issue #58619 (PR #58608). Prior history: #56576 (expired unshipped), repo-memory flagged_items.md.
Related to #56849
Warning
Firewall blocked 1 domain
The following domain was blocked by the firewall during workflow execution:
api.anthropic.com
To allow these domains, add them to the network.allowed list in your workflow frontmatter:
Description
The "policy-driven decline reported as hard failure" bug in safe-outputs (
push_to_pull_request_branchandapprove_workflow_run) has been proposed-but-unshipped since 2026-08-28. Today it hit a real production PR for the first time: run 33921923497 on PR #58608 (Design Decision Gate) failed — including on retry (run_attempt 2) — because the changeset touched files outside the workflow'sallowed-fileslist. That decline is correctly detected but surfaced through the same hard-failure path (auto-filed as #58619, expiring today) as a genuine error, instead of a distinct "declined by policy" outcome.This is the 6th occurrence of the pattern for Design Decision Gate alone (prior 5 were on internal/tooling PRs only), per today's Safe Output Health Monitor (discussion #58731). The original tracking issue (#56576, proposed 2026-08-28) auto-expired
not_plannedwithout a code change; a related tracker (#57850) is no longer resolvable (404). Re-filing to restore a live tracker, now with confirmed production impact.Expected Impact
Prevents legitimate PRs from being blocked/retried by a false-failure signal, and stops this recurring pattern from polluting safe-output failure-rate metrics across every workflow type (also affects
approve_workflow_run's permission-scope failures, which share the same root classification gap).Suggested Fix
Reclassify policy-driven declines (allowed-files, protected-files) as a distinct non-failure/soft-skip outcome across all safe-output types, not just where partially handled today — likely in the shared safe-outputs processing path (
pkg/workflowsafe_outputs job /push_to_pull_request_branchhandler).Suggested Agent
Copilot SWE Agent or any coding agent with
pkg/workflowwrite access.Estimated Effort
Medium (1-4 hours) — touches shared safe-output result classification, needs care to not swallow genuine failures.
Data Source
DeepReport analysis 2026-09-05, sourced from Safe Output Health Monitor discussion #58731 (Cluster 1) and Design Decision Gate failure issue #58619 (PR #58608). Prior history: #56576 (expired unshipped), repo-memory
flagged_items.md.Related to #56849
Warning
Firewall blocked 1 domain
The following domain was blocked by the firewall during workflow execution:
api.anthropic.comTo allow these domains, add them to the
network.allowedlist in your workflow frontmatter:See Network Configuration for more information.