DeepReport Intelligence Briefing - 2026-09-21

[github-actions[bot]]

🔍 Executive Summary

This cycle performed a full (non-incremental) analysis after repo-memory showed no updates since 2026-09-08, itself evidence of a persistence bug. Across 19 recent discussions, 500 weekly issues, and 30 workflow-run logs (4.7h runtime, ~1.16M tokens, 10 errors), the fleet is healthy overall (0 issues open >7 days, only 1 run classified "risky" and it verified as a false positive), but several concrete, fixable gaps surfaced: a reproducing repo-memory write bug despite a prior "not planned" closure, a drifted linter-registration/CI-flags pairing, a handful of documentation gaps around WIF auth and CLI install ordering, an unregistered skill file, and real bugs hiding unlabeled in a backlog dominated by auto-generated WIP stubs.

🚨 Top 5 Findings

• Repo-memory persistence bug still reproduces: all deep-report/*.md memory files remain frozen at 2026-09-08 content despite this and prior workflow runs reporting successful push_repo_memory calls; the prior tracking issue ([deep-report] push_repo_memory silently deletes repo-memory checkout while reporting false success #60773) was closed not_planned on 2026-09-16, so the bug currently has no live tracker.
• Linter registry (72 analyzers) has drifted from CI's hand-typed LINTER_FLAGS: pkg/linters/registry.go registers 72 analyzers via All(), but .github/workflows/cgo.yml hand-maintains two separate flag strings (native ~59, WASM ~58) with no automated check keeping them in sync — a silent regression risk each time a linter is added.
• Claude Code auth docs omit the WIF option that exists: docs/src/content/docs/introduction/how-they-work.mdx lists an Anthropic WIF option for Claude Code, but docs/src/content/docs/setup/quick-start.mdx's Claude section only documents ANTHROPIC_API_KEY, unlike the Gemini section which inlines its WIF pointer.
• .claude/skills/reporting.md is missing required name: frontmatter, so it silently fails to register as an invocable skill despite containing real, actively-referenced formatting guidance.
• 5 genuine bug reports are unlabeled in the weekly issues snapshot (Copilot CLI 1.0.85 (gh-aw v0.89.17) returns instant HTTP 400 on first request, breaking agent runs #62363, engine.config for the codex engine always produces an invalid lock file (injected at column 0 inside an indented run: block) #62204, claude engine: base-folder restore deletes APM-installed .claude/skills before the agent starts #62167, GH_AW_CI_TRIGGER_TOKEN still emitted in compiled lock file even when PAT is no longer required #62045, Docs: Gallery "Dependency Analysis" link points to ResearchPlanAssignOps #62076), obscured by a much larger volume of auto-generated, unlabeled [WIP] ... work in progress stub issues — making backlog triage by label unreliable.

✅ Actionable Agentic Tasks

1. Investigate and fix the reproducing repo-memory persistence gap (post-Docker-migration, post-[deep-report] push_repo_memory silently deletes repo-memory checkout while reporting false success #60773-closure).
2. Add a CI check that keeps pkg/linters/registry.go's analyzer count in sync with cgo.yml's LINTER_FLAGS strings.
3. Document what Quick Start Step 2's external repo-status workflow actually does before users run it.
4. Collapse advanced install options below the Day-one commands table in cli.md for readability.
5. Add an inline Anthropic WIF pointer to the Claude section of quick-start.mdx.
6. Add the missing name: frontmatter field to .claude/skills/reporting.md so it registers.
7. Label the 5 genuine bug reports hidden in the unlabeled issue backlog (Copilot CLI 1.0.85 (gh-aw v0.89.17) returns instant HTTP 400 on first request, breaking agent runs #62363, engine.config for the codex engine always produces an invalid lock file (injected at column 0 inside an indented run: block) #62204, claude engine: base-folder restore deletes APM-installed .claude/skills before the agent starts #62167, GH_AW_CI_TRIGGER_TOKEN still emitted in compiled lock file even when PAT is no longer required #62045, Docs: Gallery "Dependency Analysis" link points to ResearchPlanAssignOps #62076).

Methodology and data sources

• Discussions: /tmp/gh-aw/agent/discussions-data/discussions.json (100 fetched, 19 in the last 7 days, full bodies reviewed).
• Weekly issues: /tmp/gh-aw/agent/weekly-issues-data/issues.json (500 fetched; 111 open / 389 closed; top labels agentic-workflows (333), cascade-suspected (177), automation (26)).
• Workflow logs: 30 most recent runs analyzed (19 baseline / 10 normal / 1 risky-then-verified-false-positive), 9 driver-exit failures and 1 agent-logic failure noted but not newly actionable this cycle (chronic/already-tracked per repo-memory history).
• Each of the 7 tasks above was independently live-verified against current repo source (not just prior discussion text) and passed a dedup gate against existing open/recently-closed issues before filing.

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:

network:
  allowed:
    - defaults
    - "api.anthropic.com"

See Network Configuration for more information.

Generated by 🔬 Deep Report · claude · agent · 405.3 AIC · ⌖ 10.7 AIC · ⊞ 13K · ◷

• expires on Sep 28, 2026, 10:49 AM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Deep Report.

A newer discussion is available at Discussion #62508.