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
daily-experiment-report.md (.github/workflows/daily-experiment-report.md:20-22) declares network.allowed: [defaults, go] only. Its core capability, the gh aw experiments analyze CLI, needs outbound access to github.com (via git/gh) to fetch each experiments/<id> branch's decision-engine output (decision, reason_code, readiness, 30-run outcome correlation). Because github.com is not in the allowlist, every run of this workflow falls back to a descriptive-only report (sample counts + chi-square balance only) instead of computing real PROMOTE/EXTEND/REJECT/INCONCLUSIVE decisions — confirmed today (2026-09-04, run 33850615177) via the discussion's own explicit note: "gh aw experiments analyze could not run in this environment (outbound git access to github.com is proxy-blocked)" and a Firewall blocked 1 domain: github.com warning banner. This is the same failure class already fixed for other workflows by adding the github ecosystem network preset (see pkg/workflow/data/ecosystem_domains*.json, "github" preset includes github.com), e.g. issue #58184 (PR Code Quality Reviewer).
Expected Impact
Restores the workflow's actual mission (computing real experiment decisions with guardrail pass/fail) instead of a permanently degraded descriptive-only fallback that has apparently been in place for at least the last several daily runs.
Suggested Agent
New Agent (or general-purpose code-fix agent) — one-line frontmatter change: add github to network.allowed in daily-experiment-report.md, then recompile the lock file.
Description
daily-experiment-report.md(.github/workflows/daily-experiment-report.md:20-22) declaresnetwork.allowed: [defaults, go]only. Its core capability, thegh aw experiments analyzeCLI, needs outbound access togithub.com(viagit/gh) to fetch eachexperiments/<id>branch's decision-engine output (decision,reason_code,readiness, 30-run outcome correlation). Becausegithub.comis not in the allowlist, every run of this workflow falls back to a descriptive-only report (sample counts + chi-square balance only) instead of computing real PROMOTE/EXTEND/REJECT/INCONCLUSIVE decisions — confirmed today (2026-09-04, run 33850615177) via the discussion's own explicit note: "gh aw experiments analyze could not run in this environment (outbound git access to github.com is proxy-blocked)" and aFirewall blocked 1 domain: github.comwarning banner. This is the same failure class already fixed for other workflows by adding thegithubecosystem network preset (seepkg/workflow/data/ecosystem_domains*.json,"github"preset includesgithub.com), e.g. issue #58184 (PR Code Quality Reviewer).Expected Impact
Restores the workflow's actual mission (computing real experiment decisions with guardrail pass/fail) instead of a permanently degraded descriptive-only fallback that has apparently been in place for at least the last several daily runs.
Suggested Agent
New Agent (or general-purpose code-fix agent) — one-line frontmatter change: add
githubtonetwork.allowedindaily-experiment-report.md, then recompile the lock file.Estimated Effort
Quick (< 1 hour)
Data Source
DeepReport Intelligence analysis, 2026-09-04 cycle, derived from Daily Experiment Report discussion #58431.
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.