I Rebuilt BMD Desktop Around Queues and Drain

I rebuilt BMD Desktop around queues and drain on 2026-10-14. Broken overnight jobs clogged my desk. I spent the early morning cleaning up crashes, tracking unrotated keys, and merging pull request #600 to keep card actions safe. You can check what your fleet actually executes with BMD, see what your agents actually did.

Why did the overnight runs fail on 2026-10-14?

Four worker runs broke or stalled overnight because unhandled errors crashed the tasks. A digest run missed its completion gate, an audit validation exited with an error, a knowledge graph reindex hit a stack overflow, and an RSS reader dropped feeds. Silent failures stopped the fleet before sunrise on 2026-10-14.

The breakages hit early on the morning of 2026-10-14. At 05:30, the digest agent missed its completion gate. At 06:50, the validation script crashed with python exit 1. The kg-reindex run threw a stack overflow at 23:30. Later, the rss-ingestor stalled on five feeds, reading only 21 of its 26 sources.

The scheduled sweeps did not run cleanly either. Sweeps skipped at 02:40 and 04:45 on 2026-10-14 because the report for 2026-10-13 was already finalized on demand. That left merged, held, and escalated counts unverified. A separate GitHub sweep held 10 pull requests, merged zero, and classified 39 issues without touching the queue.

The brain worker processed 47 cards on 2026-10-14. It shipped 3. It sent 2 to patrick/. It blocked 5, archived 16, and skipped 21. My release budget held until 2026-10-16. If your agents stall in loops, look into AI agent cost control.

How did drain autonomy cut 63 requests down to 10?

I merged pull request #599 into bmd-desktop on 2026-10-14 to let the drain routine recommend actions without waiting for manual clicks. The automated logic cleared the backlog from 63 open requests down to 10. That change let the fleet process stalled cards while skipping my active personal work.

PR #599 merged as commit a08886a3 on 2026-10-14. It gave the drain loop recommendation autonomy. The queue dropped from 63 open requests down to 10 fast. PR #600 followed immediately as efba7d61 for version 3.47.2438, locking down status counts and card action safety after BMD-166 to BMD-168 QA passed green CI.

Earlier on 2026-10-14, PRs #595, #596, and #597 established version 3.47.2435. My profile now opens straight to Queues. A Stuck/CI/Brain/Missed strip sits across the top. The Status tab drains, queues, or resumes one card at a time, and it skips my cards entirely. PR #601 landed as 4c260fce on 2026-10-14 with loop redesign cards BMD-171 to BMD-177.

Other pieces shipped on 2026-10-14 as well. Autotrader PR #132 merged as 90884c4c for Robinhood options scanning. The main site took PR #2005 as 41fb5338 for bmdpat.com/tools/resume-osint. Four cards closed directly to Complete/: three brain items and one bmdpat-desktop task.

What did the green security report hide?

The security scan reported green with zero high-severity issues on 2026-10-14, but every underlying check actually failed. Gitleaks checked zero of five repositories because every run crashed. A dependency audit showed 27 vulnerabilities, and an unrotated credential warning sat in logs for 38 days. The green badge masked total blind spots.

SecurityAnalyst returned green on 2026-10-14: zero P0, zero P1, and 14 P2 items. That green was hollow. Gitleaks scanned zero of five repositories. All five failed. The dependency audit came back red with five critical findings and 22 high alerts.

GR-110 stayed red on 2026-10-14. It flags 18 credential-shaped strings across six runner logs. It has stayed open and unrotated for 38 days. If you test models on bare metal, verify what runs with my 5090 local inference guide.

On 2026-10-14, the CFO report called the Day 90 read: seven red, three amber, zero green. Every cell with a live number missed, and every cell that is not red has no measurement behind it. The autotrader miss is a measurement failure, not a loss. Different problem than the spend rows. Distribution stays paused and the desk count waits for 2026-10-30, so no outward push was due.

Two brain cards wait for my call on 2026-10-15: the bmd-research ARCHITECTURE.md scope, and two dated source disputes in the nightshift semantic review. The quarter-end review for 2026-10-15 is already drafted in Reports/Quarterly/2026-10-15-quarter-end.md.

What should you do with this?

1. Inspect the exit code of every scanner step before you accept a green CI summary badge.
2. Add recommendation autonomy to your queue worker to drain stalled backlogs without manual confirmation.
3. Halt new agent release budgets until unresolved red dependency alerts clear completely.

Accompanying prompt

What the prompt does: This prompt audits your queue drain and flags silent tool failures across your agent runner logs.

Copy/paste this prompt:

Copy-ready prompt

Paste the exact block into your coding agent.

No article chrome, no footnotes, no formatting drift.

Copy promptSelect text

Role: Systems Auditor Context: An automated fleet completed overnight tasks, but pipeline gates failed silently and open requests piled up in the backlog. Inputs: - Queue request count before drain: __ - Queue request count after drain: __ - Failed runner logs: __ - Gitleaks exit status: __ - Dependency audit high and critical count: __ Task: 1. Parse the failed runner logs and extract every nonzero exit code. 2. Compare the queue request count before and after the drain pass. 3. Verify if security green badges match actual zero-finding scanner passes. 4. Flag any credential-shaped occurrences that remain unrotated. Output: - Table of failed runs with file paths and exit codes - Drain efficiency ratio - List of unrotated secret warnings Constraints: - Rely only on supplied log values. - Do not assume a scan passed if zero files were analyzed.

27 lines/859 chars

Ready

This prompt and every other one we publish live in the free prompt library.

Copy the block above.

Weekly measured local runs: https://bmdpat.com/5090-reports