DeepReport Intelligence Briefing - 2026-10-05

[github-actions[bot]]

🔍 Executive Summary

Narrow incremental cycle (~5.7h since prior briefing #65643): only 4 new discussions, 3 healthy/informational and 1 total report failure. The notable finding is DeepReport's own, root-caused chronic bug: unscoped agenticworkflows logs fleet-wide queries hit a 5-minute server-side timeout floor that the agent's typical ~50-60s wait budget never clears — this has silently recurred across many prior cycles but was never filed as its own issue until now. The previously-filed push_repo_memory nested-subdirectory glob bug (#65642) is confirmed still unfixed; this cycle worked around it directly to restore memory persistence. Repository health elsewhere is unremarkable and stable.

🚨 Top 5 Findings

• Root-caused and filed: agenticworkflows logs calls without a workflow_name filter hit a 5-minute server-side timeout floor (pkg/cli/mcp_tools_privileged.go: defaultMCPLogsMinTimeoutMinutesAllWorkflows=5, defaultMCPLogsToolCount=100), but deep-report.md's Step 2 gives no scoping guidance and the agent's typical foreground wait budget (~50-60s) is far shorter — reproduced live this cycle (6h-window call returned only keepalive pings, no result), consistent with the same timeout noted in at least 2 prior cycles' memory. Filed as a new issue (first time as its own bug rather than worked around).
• Confirmed still open, not re-filed: push_repo_memory nested-subdirectory file-glob bug ([deep-report] push_repo_memory silently drops nested-subdirectory memory files (6 workflows affected, incl. deep-report's own 26 [Content trun #65642, tracked under AW Top 10 [AW Top 10] 09 Fix repo-memory file-glob dropping nested directory files #65657) — verified deep-report.md:73's file-glob is still slashless (["*.md", "*.json"]). This explains why this very workflow's repo-memory was stuck at 2026-09-08 for ~26 days. Workaround applied this cycle: wrote flat, depth-0 deep-report-*.md copies at the repo-memory root (which the current slashless glob does match), confirmed via push_repo_memory's own validator reporting 3 real files / 23 KB this time (vs. "0 files" every prior cycle) — memory persistence is restored pending the real fix.
• Watched, not filed (single occurrence, no cross-confirmation): Daily Team Evolution Insights ([daily-team-evolution] Daily Team Evolution Insights - 2026-10-04 #65663) produced a completely empty report — "GitHub MCP tools are not available in the current environment" — despite github: mode: local being configured in its frontmatter. It's the only goose-engine workflow using mode: local GitHub tools in the fleet, so there's no sibling workflow to confirm whether this is a recipe-wide incompatibility or a one-off MCP gateway hiccup.
• Watched, not filed (root cause unconfirmed): Daily Observability Report ([observability] Observability Coverage Report - 2026-10-04 #65685) reported "Permission denied" traversing /tmp/gh-aw/aw-mcp/logs for per-run telemetry file validation. Investigated pkg/cli/mcp_logs_guardrail.go's nearby comment, but it describes a different cache directory's artifact-upload-exclusion design, not a documented restriction on this path — inconclusive on one occurrence.
• Healthy: Lockfile Statistics ([lockfile-stats] Lockfile Statistics Audit — 2026-10-04 #65664, 320 lockfiles, stable) and Copilot PR Prompt Pattern Analysis ([prompt-analysis] Copilot PR Prompt Analysis - 2026-10-04 #65669, 76.3% merge rate, narrow-scope-wins insight) are both routine informational reports with no fileable gaps. Weekly issues snapshot: 500 issues, 143 open / 357 closed, 0 open >7 days, 60 unlabeled (chronic WIP-stub pattern).

✅ Actionable Agentic Tasks

1. Scope or budget for agenticworkflows logs's 5-minute unfiltered-query timeout floor in deep-report.md Step 2 (filed as a GitHub issue this cycle) — the single highest-leverage fix available this window: it unblocks DeepReport's own fleet-health checks, which have been silently skipped nearly every cycle in favor of indirect discussion-report signals.

No further candidates cleared the verification bar this narrow ~5.7h window: the Team Evolution and Observability findings are each single-occurrence with no confirmed root cause or cross-workflow signal (watched for a 2nd occurrence before filing); the memory glob bug and cache-memory gap are already tracked under #65642/#65657/#65641; and #65173 (a real human-reported bug flagged in the prior cycle) remains blocked by an integrity-policy filter this cycle too. Filing fewer, well-verified tasks rather than padding to a quota.

View Full Details

agenticworkflows logs timeout — reproduction and root cause

• Live reproduction: agenticworkflows logs --start-date -6h (no workflow_name) returned only MCP keepalive pings for ~50s with no final tool result.
• pkg/cli/mcp_tools_privileged.go: effectiveMCPLogsToolTimeoutMinutes applies defaultMCPLogsMinTimeoutMinutesAllWorkflows = 5 whenever no workflow_name is given (or an engine filter is used instead), on top of a default count of 100 runs — by design, since an unfiltered query scans every workflow and reads its artifacts.
• deep-report.md's "Step 2: Gather Workflow Intelligence" instructs using the logs tool with no mention of scoping by workflow or count, so cycles typically call it unscoped and give up well before the 5-minute floor the tool itself expects to need.
• This matches prior cycles' memory notes recording the identical symptom (e.g. the 2026-09-07T~18:32Z cycle and immediately-prior DeepReport Intelligence Briefing - 2026-10-04 (cycle 3) #65643), which each worked around it rather than filing a fix — this cycle filed it directly.

push_repo_memory glob bug — confirmed still open, workaround applied

• deep-report.md:73 file-glob: ["*.md", "*.json"] is slashless; deep-report's prompt saves all memory files under a nested deep-report/ subdirectory, so every file's relative path fails the depth-0 match and push_repo_memory silently reports "0 eligible files."
• This cycle, in addition to the nested files, wrote flat depth-0 copies (deep-report-last_analysis_timestamp.md, deep-report-flagged_items.md, deep-report-processed-discussions.md) directly under /tmp/gh-aw/repo-memory/default/. push_repo_memory's validator confirmed 3 files / 23 KB accepted — first successful memory persistence in ~26 days of cycles. Future cycles should check the flat deep-report-*.md files first until the underlying glob-eligibility bug ([deep-report] push_repo_memory silently drops nested-subdirectory memory files (6 workflows affected, incl. deep-report's own 26 [Content trun #65642 / AW Top 10 [AW Top 10] 09 Fix repo-memory file-glob dropping nested directory files #65657) is actually fixed.

Daily Team Evolution Insights total failure

• Full report body: Error encountered while gathering repo activity data: GitHub MCP tools are not available in the current environment.
• Frontmatter has tools: github: { mode: local, toolsets: [repos, issues, pull_requests, discussions] } and engine: { id: goose }, model: copilot/auto.
• It's the only workflow in the fleet combining goose + github: mode: local, so there's no sibling run to confirm whether this is a recipe-level incompatibility or a transient MCP gateway issue. No matching prior occurrence found in the last 100 discussions. Watching for a 2nd occurrence before filing.

Daily Observability Report permission-denied warning

• Report states per-run recursive file validation for access.log/gateway.jsonl/rpc-messages.jsonl was blocked by "Permission denied" traversing /tmp/gh-aw/aw-mcp/logs/run-<id> directories, downgrading findings to inconclusive warnings rather than confirmed coverage gaps.
• Checked pkg/cli/mcp_logs_guardrail.go for a related permission-handling comment; it explains why MCP summary JSON files live in a separate /tmp/gh-aw/logs-cache directory (to keep them out of artifact uploads) — an upload-scoping design choice, not a stated access restriction on /tmp/gh-aw/aw-mcp/logs itself. Root cause unconfirmed on this one occurrence; watching.

Weekly issues snapshot

500 issues analyzed: 143 open (28.6%) / 357 closed, 0 open >7 days, 60 unlabeled (chronic WIP-stub auto-pattern). Top labels: agentic-workflows (294), automation (116), cascade-suspected (82), cookie (74), testing (40). Top authors: app/github-actions (483, dominant bot-authored volume), niebloomj (7), MureedSultan (2).

Dedup checks performed

• agenticworkflows logs timeout fix: searched local weekly-issues-data/issues.json for logs.*timeout|timeout.*logs|agenticworkflows.*timeout — no match, filed.
• Confirmed via local issues snapshot that push_repo_memory/glob ([deep-report] push_repo_memory silently drops nested-subdirectory memory files (6 workflows affected, incl. deep-report's own 26 [Content trun #65642, [AW Top 10] 09 Fix repo-memory file-glob dropping nested directory files #65657) and daily-secrets-analysis cache-memory ([deep-report] Re-file: daily-secrets-analysis still missing cache-memory (prior issue #64554 auto-expired unaddressed) #65641) are already open and unresolved — not re-filed.
• protected-files exclude unprotects a file name at every depth: no way to allow only the root pyproject.toml #65173 (protected-files.exclude depth bug, flagged by the prior cycle) remains inaccessible this cycle too (issue:github/gh-aw#65173 filtered by integrity policy, "integrity below approved") — left untouched again.

Generated by 🔬 Deep Report · claude · agent · 216.9 AIC · ⌖ 8.14 AIC · ⊞ 7.1K · ◷

• expires on Oct 11, 2026, 5:18 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Deep Report.

A newer discussion is available at Discussion #65800.