v0.89.8

🌟 Release Highlights

This release strengthens billing and guardrail accuracy for AI Credits, hardens safe-output permission enforcement, and adds new configuration flexibility for OTLP telemetry and agent assignment.

✨ What's New

• Secret-backed OTLP endpoints — Organization-wide OpenTelemetry endpoints can now be configured via secrets.GH_AW_DEFAULT_OTLP_ENDPOINT in addition to visible variables, with automatic fallback and existing precedence preserved.
• Reasoning effort for assign-to-agent — Forward model-specific reasoning_effort (e.g. high) through agent assignment, with automatic validation and graceful fallback for unsupported agents/models.
• continue-on-error on generated agent jobs — The compiler now correctly preserves an explicit continue-on-error: true/false on the generated agent job instead of silently dropping it.
• Date-bounded cached JSONL logs — gh aw logs --cached-jsonl now prunes cached run records outside the requested --start-date/--end-date range instead of only appending, keeping cached history accurate.
• Centralized slash commands get PR read access — pre_activation now automatically receives pull-requests: read when a workflow defines a centralized slash command, fixing silent execution skips on pull requests.

🐛 Bug Fixes & Improvements

• Fixed the daily AI Credits gate incorrectly rejecting failed agent runs that had authoritative zero-credit usage artifacts.
• Fixed the daily AIC guardrail treating pre-agent setup failures (with no model usage) as potentially billable, by tracking explicit execution-started/not-started evidence.
• Hardened add_labels to enforce configured targets, preventing labels from being applied outside the triggering issue or PR unless explicitly allowed via target: "*".
• Prevented benign tool-result text (e.g. "not logged in", EACCES) in Claude stream-JSON output from incorrectly triggering failure classifiers and suppressing retries.

📚 Documentation

• Tightened the trace graders documentation for clarity and conciseness.

Generated by 🚀 Release · copilot · auto · 19.6 AIC · ⊞ 11.9K

---

What's Changed

• Fix dangling determine-automatic-lockdown step reference in static enclave sink-visibility policy by @lpcox with @Copilot in #60338
• [docs] docs: unbloat trace graders doc by @github-actions[bot] in #60359
• Grant PR read permission to centralized slash commands by @pelikhan with @Copilot in #60347
• Prevent Claude tool results from triggering failure classifiers by @pelikhan with @Copilot in #60349
• Support secret-backed default OTLP endpoints by @pelikhan with @Copilot in #60348
• Retain cached JSONL history within requested date ranges by @pelikhan with @Copilot in #60361
• Preserve continue-on-error on generated agent jobs by @pelikhan with @Copilot in #60350
• Add reasoning effort support to assign-to-agent by @pelikhan with @Copilot in #60346
• Prevent daily AIC guardrail blocks after pre-agent failures by @lpcox with @Copilot in #60355
• Enforce add-labels target restrictions by @pelikhan with @Copilot in #60362
• Fix daily AI Credits gate accounting fallback by @pelikhan with @Copilot in #60371

Full Changelog: v0.89.7...v0.89.8