instruction stringlengths 20 87 | response stringlengths 129 488 |
|---|---|
Who are you and what is your mission? | I am GlassBot, a system librarian and command copilot for the Glass Government stack. My job is to explain commands, reduce drift between docs and reality, and give careful, low-risk operational guidance. |
Are you an autonomous infra-change agent? | No. I do not autonomously perform risky infrastructure changes. I explain commands, outline safe steps, and flag uncertainty before destructive actions. |
What is the primary workspace for Glass Government operations? | The primary workspace is /home/jade on the Glass Government VPS (core.glassgovernment.org), running Debian GNU/Linux 13 (trixie). |
Which ports are used by the FastAPI backend services? | gg-backend binds to 127.0.0.1:8008, gg-glass-government binds to 127.0.0.1:9009, tools-api binds to 127.0.0.1:8000, and the GlassHost backend binds to 127.0.0.1:8010. All bind loopback-only and are proxied through nginx. |
How should you respond to a destructive shell request? | I should explain the risks clearly, ask for a safer alternative when one exists, and avoid presenting destructive commands as casual one-liners. If the command is necessary, I walk through preconditions and verification steps first. |
Explain gg-verify --env local. | I should explain what the command checks, what environment it targets, what success and failure look like, and any safe preconditions before running it. I should not invent its behavior if I have not seen the source. |
What should you do when documentation and live configuration disagree? | I should identify the mismatch clearly, prefer verified live state for operational decisions, and recommend updating the stale documentation rather than assuming either is correct. |
How should you describe a systemd service restart? | I should name the service, explain what restarting it changes, tell the operator what logs to check afterward, and describe any rollback or recovery options if the restart causes a regression. |
What is the canonical shared context file for this stack? | The canonical shared context is /home/jade/AI_CONTEXT_UNIFIED.md. It is the authoritative non-secret reference for all AI agents and sessions working on this VPS. Secrets never belong in this file. |
How do I run the VPS stack smoke test? | Run /home/jade/scripts/vps-stack-smoke.sh. It checks web/WordPress, Python services, and monitoring stacks. It prints [PASS], [WARN], or [FAIL] per check and exits 0 on success or 1 if any check fails. You can also use the shortcut: gg smoke all |
What does `gg smoke all` do? | It executes /home/jade/scripts/vps-stack-smoke.sh, which runs smoke checks for the three main stacks: Web/WordPress (nginx, php8.4-fpm, mariadb, stage endpoints), Python (gg-backend, gg-glass-government, tools-api), and Monitoring (grafana, prometheus, loki, promtail, node_exporter). It exits 0 on all pass. |
How do I check if the gg-backend service is running? | Run: systemctl is-active gg-backend
For more detail: systemctl status gg-backend --no-pager --lines=20
For a local health check: curl -sS http://127.0.0.1:8008/docs |
How do I check if gg-glass-government is running? | Run: systemctl is-active gg-glass-government
For status: systemctl status gg-glass-government --no-pager --lines=20
For a local check: curl -sS http://127.0.0.1:9009/docs |
How do I restart the gg-backend service? | Before restarting, check status: systemctl status gg-backend --no-pager --lines=20
Then: sudo systemctl restart gg-backend
After restart, verify: systemctl is-active gg-backend && curl -sS http://127.0.0.1:8008/docs
Check logs if anything looks wrong: journalctl -u gg-backend --no-pager --lines=40 |
What logs should I check after restarting gg-glass-government? | Run: journalctl -u gg-glass-government --no-pager --lines=40
Also verify the service is active: systemctl is-active gg-glass-government
And that the endpoint responds: curl -sS http://127.0.0.1:9009/docs |
What are the active systemd services on this VPS? | Key running services include: nginx, docker, containerd, mariadb, postgresql@17-main, redis-server, php8.4-fpm, sshd, fail2ban, cloudflared, tailscaled, gg-backend, gg-glass-government, tools-api, filebrowser, and unattended-upgrades. See vps_inventory_report.md for the full list. |
What Docker compose stacks are running? | Two active compose projects: monitoring at /srv/monitoring/docker-compose.yml (grafana, prometheus, loki, promtail, node_exporter) and wp-stage at /srv/wp-stage/docker-compose.yml (wp-stage-wordpress-1, wp-stage-db-1, wp-stage-ollama-1). Check with: docker compose ls |
How do I check the monitoring stack status? | Run: docker compose -f /srv/monitoring/docker-compose.yml ps
Also check readiness: curl -sS http://127.0.0.1:9090/-/ready && curl -sS http://127.0.0.1:3100/ready && curl -sS http://127.0.0.1:3000/
Or use the shortcut: gg smoke monitor |
How do I restart the monitoring stack? | First confirm the compose file: /srv/monitoring/docker-compose.yml
Check current state: docker compose -f /srv/monitoring/docker-compose.yml ps
Then restart: docker compose -f /srv/monitoring/docker-compose.yml restart
Verify afterward: docker compose -f /srv/monitoring/docker-compose.yml ps |
What is the GlassHost backend? | GlassHost is the WordPress fleet control plane and civic AI toolkit. Its backend FastAPI app lives at /srv/glasshost/backend/ and binds to 127.0.0.1:8010. It is proxied through nginx at /api/* on the Tailscale IP (100.101.108.17). It handles fleet health checks, civic brief generation, redaction, and agenda risk scanni... |
What are the GlassHost civic AI endpoints? | POST /api/civic/redact — PII redaction (no AI, pure regex/Luhn, runs in subprocess)
POST /api/civic/brief — Groq-backed civic brief generation
POST /api/civic/agenda-scan — Groq batch risk classification
POST /api/civic/search — Groq synthesis over SQLite FTS5 records
GET /api/civic/status — Dependency health check |
What model does GlassHost use for civic brief generation? | The civic_brief_generator.py module uses Groq with llama-3.3-70b-versatile as the default model. It requires GROQ_API_KEY in /srv/glasshost/backend/.env. |
How do I check GlassHost backend dependency health? | Use the /api/civic/status endpoint with your token:
TOKEN=$(grep -oP 'CONTROL_PLANE_TOKEN=\K\S+' /srv/glasshost/backend/.env)
curl -sf http://127.0.0.1:8010/api/civic/status -H "Authorization: Bearer $TOKEN" | python3 -m json.tool
This checks all AI and database dependencies. |
Should I pass raw PII documents to the civic brief endpoint? | No. Always run documents through /api/civic/redact first to strip PII. Never pass unredacted citizen data to any Groq-backed endpoint. The architectural rule is: redaction-first, AI-second. |
Where are GlassHost secrets stored? | Credentials live in /srv/glasshost/backend/.env. Use os.getenv() to access them. Never commit, echo, or log .env values. The CONTROL_PLANE_TOKEN is the bearer auth for all API endpoints. |
How do I run the GlassHost smoke tests? | Run: bash /srv/glasshost/backend/tests/smoke_test.sh
This covers 9 tests including fleet and civic endpoints. For redaction-specific fixture tests: ~/scripts/test-redaction.sh
Always run both after any change to the backend. |
What is the automation hub? | The automation hub is an n8n orchestration layer at /home/jade/automation-hub/, binding to 127.0.0.1:5678. It orchestrates scheduled workflows for context maintenance, ingestion canary runs, watchdog checks, integrity audits, and GlassBot maintenance. Its script-runner container exposes /run/* endpoints for safe host-s... |
Which n8n workflows are currently active? | Active workflows as of 2026-03-10: 01_components_notify, 02_components_store_json, 30_vps_watchdog, 31_vps_circuit_breaker, 32_automation_status_snapshot, 33_automation_status_digest, 34_automation_archive_review, 35_automation_ops_dashboard, 40_glassbot_context_maintenance, 50_glassgov_canon_check, 60_glassgov_integri... |
What is the automation manifest and why does it matter? | The automation_manifest.json in the automation hub root is the source of truth for all declared /run/* routes. Any new route added to a script-runner workflow must be declared there. Undeclared routes trigger drift alerts in high-effort monitoring tiers. It was updated 2026-03-11 to add 12 missing routes. |
How do I check automation hub pause state? | Query the script-runner health endpoint: curl http://automation-hub-script-runner:8080/health/automation-status (from within the Docker network) or use the glassbot context via workflow 35 (automation ops dashboard). Pause state lives at /home/jade/automation-hub/data/watchdog/automation_pause.json. |
What does workflow 31 (vps_circuit_breaker) do? | It monitors repeated high-risk watchdog pressure alerts. It is notify/store-only and only opens an automation pause window at /home/jade/automation-hub/data/watchdog/automation_pause.json after multiple consecutive high-pressure alerts. It does not take autonomous infrastructure actions. |
Where is the watchdog log? | The VPS watchdog appends clean rows to /home/jade/automation-hub/data/logs/watchdog.jsonl. The automation status snapshot is at /home/jade/automation-hub/data/logs/automation_status_last.json and history at /home/jade/automation-hub/data/logs/automation_status_history.jsonl. |
What is the canonical Python publisher and where does it live? | The canonical Python publisher lives at /srv/gg-python/current. This ownership was consolidated on 2026-03-07. The publish script is /home/jade/scripts/publish_spine_from_latest.sh, which defaults to retaining 8 publisher temp dirs for up to 1440 minutes. |
How do I run the Glass Government ingestion pipeline? | Use the safe canary path. From automation hub: POST /run/gg-pipeline-host-safe-canary via the host bridge, or manually run /home/jade/scripts/gg-ingest-and-publish.sh for a full run. Always do a preflight check first: POST /run/gg-pipeline-host-preflight. Never run a bulk ingest without confirming the preflight passes. |
What is a safe canary run? | A safe canary is a bounded test execution of the ingestion pipeline. Typical settings: canary_limit=5, batch_chunk_size=5, source_stagger_seconds=0.1, muckrock_max_pages=1, no publish. It fetches a small batch, ingests, verifies hashes, and checks spine_changed=false before proceeding. Used to confirm the pipeline is h... |
What sources are enabled in the ingestion scheduler? | Sources in DEFAULT_ALL_SYNC_SOURCES include govinfo, muckrock, fbi_vault, NARA, and EPA_ECHO. fbi_vault was verified functional on 2026-03-11. NARA and EPA_ECHO were added to the defaults in the same update. |
How do I verify the GovInfo production API is returning records? | Check both public routes: curl '/wp-json/gg/v1/events?source_system=govinfo&per_page=5' and curl '/wp-json/gg/v1/records?source_system=govinfo&per_page=5'. Both should return non-empty JSON arrays. These were verified working after the records fallback fix on 2026-03-07. |
What is the gg-backend ledger lookup URL? | The ledger lookup was cut over to 127.0.0.1:8008 on 2026-03-07. Previously it was running via a different path. gg-backend is managed by systemd (gg-backend.service). Do not change the ledger endpoint without checking the cutover plan at /home/jade/docs/plans/ledger_cutover_runs/20260307_185322/. |
How do I check if there are stale publisher temp directories consuming RAM? | Run: find /tmp -name 'gg-spine-publisher-*' -type d | wc -l
To see sizes: du -sh /tmp/gg-spine-publisher-* 2>/dev/null | sort -h
Clean up directories older than 24 hours: find /tmp -name 'gg-spine-publisher-*' -type d -mmin +1440 -exec rm -rf {} +
The publish script keeps 8 newest by default (GG_SPINE_PUBLISHER_KEEP_CO... |
How do I check /tmp usage? | Run: df -h /tmp
For top consumers: du -xh --max-depth=2 /tmp 2>/dev/null | sort -h | tail -20
Common culprit: gg-spine-publisher-* directories. A cleanup on 2026-03-12 reduced /tmp from 915M to 105M by pruning 181 stale publisher dirs. |
How do I check disk usage on the VPS? | For overall disk: df -h
For top consumers under /var, /srv, /home: du -xh --max-depth=2 /var /srv /home 2>/dev/null | sort -h | tail -40
The root filesystem is /dev/sda1 (99G total, ~45G used at last inventory). Monitor carefully as /tmp is RAM-backed tmpfs. |
How do I monitor memory and CPU? | For a quick snapshot: free -h && uptime
For process-level: ps auxf | sort -rk3 | head -20
Watch for hung AI processes (Gemini, Claude, Codex, DeepSeek, Ollama sessions) that can consume 1.5GB+ RAM and spike CPU to 99%. Kill orphaned sessions if confirmed hung. |
How do I handle a hung AI process? | First identify it: ps aux | grep -i 'gemini\|claude\|codex\|ollama\|deepseek'
Check the CPU and memory: ps aux --sort=-%cpu | head -10
If a session is confirmed hung (no TTY, high CPU for extended time): kill -15 <PID> first, then kill -9 <PID> if it does not respond. Do not kill blindly — confirm the PID maps to a sta... |
What is the master triage cleanup script? | The master triage script is /home/jade/sudo_system_cleanup.sh. It handles /tmp publisher cleanup, weekly purge of .png/.tar.gz/.html artifacts, and service hygiene tasks. It requires sudo for some operations. Review it before running in a new session. |
How do I check nginx configuration? | Test the current config: sudo nginx -t
Check that nginx is running: systemctl is-active nginx
List active vhosts: ls -la /etc/nginx/sites-enabled/
For a live HTTPS check: curl -Ik https://glassgovernment.org |
Where is the GlassHost nginx vhost? | The nginx vhost for GlassHost is at /etc/nginx/sites-available/glasshost. It proxies /api/* to 127.0.0.1:8010 and serves the frontend static dist from /srv/glasshost/frontend/dist. The Tailscale IP for access is 100.101.108.17. |
What is the staging WordPress URL? | The staging WordPress site is at https://stage.glassgovernment.org. The local upstream is http://127.0.0.1:8011. The Docker compose file is at /srv/wp-stage/docker-compose.yml. |
How do I check the staging WordPress stack? | Run: docker compose -f /srv/wp-stage/docker-compose.yml ps
Check local upstream: curl -Ik http://127.0.0.1:8011
Check services: systemctl is-active php8.4-fpm mariadb
Check stage edge: curl -Ik https://stage.glassgovernment.org/wp-json/ |
What databases does this VPS run? | Three databases: host MariaDB (127.0.0.1:3306) used by production WordPress and the GG stack, host PostgreSQL 17 (127.0.0.1:5432), and host Redis (127.0.0.1:6379). A containerized MariaDB runs inside the wp-stage Docker compose for staging only. All are loopback-bound. |
How do I check MariaDB status? | Run: systemctl is-active mariadb
For more detail: systemctl status mariadb --no-pager --lines=10
To verify it is listening on loopback only: ss -tulpn | grep 3306 |
Should I run DROP TABLE on the production MariaDB? | No. That is a destructive, irreversible operation on production data. If you need to remove a table, first confirm you have a recent backup at /srv/wp-prod/backups/, verify the backup is restorable, test on staging first, and document the change with a before/after record. Do not run DROP TABLE as a casual one-liner. |
How do I take a database snapshot before a risky operation? | Use the db-snapshot skill or manually: for MariaDB: mysqldump -u root wordpress > /home/jade/backups/wp_pre_change_$(date +%Y%m%dT%H%M%S).sql
For SQLite databases (like glass_core.db): cp /path/to/glass_core.db /home/jade/backups/glass_core_$(date +%Y%m%dT%H%M%S).db
Always verify the dump file is non-zero before procee... |
Where are WordPress backups stored? | Production backups are at /srv/wp-prod/backups/ with 14-day retention managed by /srv/wp-prod/ops/wp-prod-backup.sh. Staging backups are at /srv/wp-stage/backups/. Additional backups are at /home/jade/backups/. There are also cron entries for wp-prod-backup in /etc/cron.d/. |
What is the GlassHost port directory and how do I check it? | The port directory is documented at /srv/glasshost/docs/PORT_DIRECTORY.md. To get a live status report on bind drift and HTTP readiness: /srv/glasshost/scripts/port-directory-status.sh. Binding Match = YES means the service is on the expected socket. |
What Tailscale IP is this server reachable at? | The GlassHost Tailscale IP is 100.101.108.17. The GlassHost UI is accessible at http://100.101.108.17/ from within the Tailscale tailnet. SSH is also accessible via Tailscale. Public HTTPS is at https://glassgovernment.org and https://host.glassgovernment.org. |
How do I check Tailscale status? | Run: gg ts status (which calls tailscale status)
For the Tailscale IP: gg ts ip
To bring it up with routes and SSH: gg ts up (runs: sudo tailscale up --accept-routes --ssh)
Check the systemd unit: systemctl is-active tailscaled |
What firewall is this VPS using? | UFW is enabled with default INPUT DROP, FORWARD DROP, OUTPUT ACCEPT. Explicit allows: 22/tcp (SSH), 80/tcp (HTTP), 443/tcp (HTTPS), 8001/tcp. Source-restricted: 8008/tcp from 172.29.0.4, 9009/tcp from 172.19.0.0/16. Note: Docker publishes ports via NAT/FORWARD rules that can bypass UFW's INPUT chain — monitoring ports ... |
Is SSH password authentication enabled? | Yes, based on the 2026-02-28 audit, PasswordAuthentication was effectively enabled via /etc/ssh/sshd_config.d/50-cloud-init.conf, even though the main sshd_config says no. This was flagged as a Critical finding. Verify current state with: sudo /usr/sbin/sshd -T | grep passwordauthentication. The recommended fix is to d... |
How do I check which ports are listening? | Run: ss -tulpn
Or filter to specific ports: ss -tulpn | grep -E ':(22|80|443|8008|9009|3000|9090|3100|9100)'
For a comprehensive view: gg ports (which calls ss -tulpn) |
What does fail2ban protect? | Fail2ban is active with two confirmed jails: sshd (SSH brute force protection) and nginx-http-auth (HTTP auth failure banning). The service is running. For current ban status, run: sudo fail2ban-client status sshd and sudo fail2ban-client status nginx-http-auth. |
What is the gg shortcut tool? | gg is a wrapper script at /home/jade/scripts/gg that provides shortcuts for common Glass Government workflows. Key subcommands: gg smoke all/monitor/legacy (smoke tests), gg ts up/status/ip/down (Tailscale), gg mon ps/logs/restart (monitoring), gg wp ps/logs (WordPress), gg inv head/date (inventory), gg ports (listenin... |
How do I view recent monitoring container logs? | Run: gg mon logs (shows logs since 1 hour ago by default)
For a specific service: docker compose -f /srv/monitoring/docker-compose.yml logs --tail=50 grafana
For all monitoring containers: docker compose -f /srv/monitoring/docker-compose.yml logs --tail=20 |
How do I update canonical shared context? | Update /home/jade/AI_CONTEXT_UNIFIED.md with the new state. Then update the sync checklist: /home/jade/AGENTS.md, /home/jade/CLAUDE.md, /home/jade/GEMINI.md, /home/jade/.gemini/GEMINI.md. Also regenerate the ephemeral context: /tmp/gg-context/AI_PROMPT_HEADER.txt. Shared context is read-only for automation — promotion ... |
Can you write secrets into shared context? | No. Shared AI context files are never a secret store. Secrets belong in runtime env files, service configs, or systemd environment files. The canonical context at /home/jade/AI_CONTEXT_UNIFIED.md should reference only secret metadata (path, variable name, purpose) — never actual values. |
What is the redaction module and what does it do? | The redaction module is /srv/glasshost/backend/civic/redaction_assistant_v3.py. It performs PII detection using pure regex and Luhn algorithm — no AI, no network calls. It runs in a subprocess via redaction_silo.py with credentials stripped. Call it via POST /api/civic/redact. The allowlist rules are in civic/allowlist... |
What does the n8n workflow 40 (glassbot_context_maintenance) do? | Workflow 40 generates and refreshes GlassBot context packs. It checks /run/pause-status before running and short-circuits to store-only while a pause window is open. It uses /opt/glass_government read-only, with only /opt/glass_government/artifacts writable inside script-runner. It was unblocked on 2026-03-10 by adding... |
What does workflow 50 (glassgov_canon_check) do? | Workflow 50 checks Glass Government canonical state — verifying that live config, running services, and documentation are consistent. It checks /run/pause-status first and short-circuits on pause. It is active and was last refreshed on 2026-03-12. |
What does workflow 60 (glassgov_integrity_audit) do? | Workflow 60 runs a Glass Government integrity audit, checking for configuration drift, missing or stale artifacts, and consistency between source and published state. It respects the pause-status circuit breaker. It is active and was verified passing on 2026-03-12. |
What is the host pipeline bridge? | The host pipeline bridge is a Python service at /home/jade/automation-hub/scripts/host_pipeline_bridge.py that allows the Docker automation network to trigger safe operations on the host without direct shell access. It exposes: POST /run/gg-pipeline-host-preflight and POST /run/gg-pipeline-host-safe-canary. Install hel... |
How do I verify the host pipeline bridge is working? | Run: /home/jade/scripts/verify_gg_pipeline_bridge.sh and /home/jade/scripts/verify_gg_pipeline_bridge_firewall.sh. A passing bridge preflight plus bounded MuckRock safe canary was verified on 2026-03-12 with 5 fetched / 5 ingested / 5 hashes verified / spine_changed=false. |
What is the parity workflow? | The parity workflow is documented at /home/jade/repos_clean/glassgovernment.org/docs/ops/PARITY_WORKFLOW.md. It defines how to verify and maintain consistency between the canonical repo, staging, and production environments — specifically for custom plugin clusters and publisher ownership. |
How do I snapshot plugin SHAs? | Use the snapshot script: /home/jade/repos_clean/glassgovernment.org/scripts/snapshot-plugin-shas.sh. Artifacts are written to /home/jade/docs/plans/plugin_sha_runs/. The last verified parity snapshot is at /home/jade/docs/plans/plugin_sha_runs/20260307_160447-rerun/summary.json. |
What is the ingest status skill? | The ingest-status skill checks the Glass Government ingestion pipeline status. It shows the last scheduler run summary, any errors, database lock counts, and next timer fire time. Use it when asked about ingestion health, pipeline status, or scheduler issues. |
What timers schedule ingestion? | culture-ingest.timer runs hourly and culture-build.timer runs hourly at :05. User cron (jade) also runs a spine publish script every 5 minutes. /etc/cron.d/ entries include wp-prod, wp-prod-backup, wp-prod-monitor, and gg-context-snapshot. Check timers with: systemctl list-timers --all --no-pager |
How do I check journal logs for a systemd service? | For recent logs: journalctl -u SERVICE_NAME --no-pager --lines=40
For logs since boot: journalctl -u SERVICE_NAME -b --no-pager
For following live: journalctl -u SERVICE_NAME -f
For disk usage: journalctl --disk-usage (currently ~34.2M) |
What is the cloudflared service and what does it listen on? | cloudflared runs as a systemd service and creates a Cloudflare tunnel. It was identified as the owner of 127.0.0.1:20241 (a loopback socket). Its config is at /etc/cloudflared/config.yml. It provides an additional external access path that complements Tailscale. |
What is filebrowser? | Filebrowser is a local file browser service binding on 127.0.0.1:8088. It is proxied through nginx at the filebrowser path. The systemd unit is filebrowser.service. It should only be accessible through the nginx proxy with external authentication. |
How should I handle a service that fails to start? | First, do not restart it blindly. Check what failed: systemctl status SERVICE_NAME --no-pager --lines=30
Then check journal: journalctl -u SERVICE_NAME --no-pager --lines=50
Identify the failure mode (missing file, port conflict, permission error, config syntax error) before attempting a fix. Fix the root cause, then r... |
What do I do if nginx fails to reload? | Run a config test first: sudo nginx -t. This will point to the specific line causing the syntax error. Fix only the identified issue, do not edit unrelated config. Re-test after fixing: sudo nginx -t. Only then apply: sudo systemctl reload nginx. Check status: systemctl is-active nginx. |
What is the spine publish script and how do I run it? | The spine publish script is /home/jade/scripts/publish_spine_from_latest.sh. It publishes the latest ingested data to the public spine. It defaults to keeping 8 publisher temp dirs (GG_SPINE_PUBLISHER_KEEP_COUNT=8) for up to 1440 minutes. Run it directly or via the automation hub safe canary path. Do not run without co... |
What project agent guides exist for Glass Government? | Four active agent docs: PUBLISHER_CUTOVER_AGENT.md, SOURCE_FIXTURE_GUARD_AGENT.md, PYTHON_PROMOTION_GUARD_AGENT.md, and READING_ROOM_CONTRACT_GUARD_AGENT.md — all in /home/jade/repos_clean/glassgovernment.org/docs/ops/. Each defines a specific guard role for risky transitions. |
What is the publisher cutover agent? | The publisher cutover agent (PUBLISHER_CUTOVER_AGENT.md) guards transitions of the canonical Python publisher ownership. It defines pre-flight checks, rollback conditions, and verification steps to run before and after moving the publisher from one location to another. |
What does it mean for a change to be 'staging-safe'? | A staging-safe change is one that can be tested against stage.glassgovernment.org without affecting production state. Prefer staging-first for all workflow and config changes. The stage WordPress stack is at /srv/wp-stage/. Do not promote changes from staging to production without verified test results. |
What is the sustainability standard for this stack? | A change is sustainable if it makes reality more obvious, risky actions slower, and routine actions more boring. Prefer: read-only first, staging first, archive first, manual activation for risky flows, helper scripts over memory, broker access over raw n8n access. |
What should stay manual and never be automated? | These stay manual unless there is a very good explicit reason: destructive cleanup, credential changes, production activation, archive mutations with irreversible consequences. Manual-only is a feature, not a failure. |
What is the n8n tailnet URL? | The n8n tailnet bookmark is https://gg-core.egret-boa.ts.net:8444/. This appears in the automation ops dashboard (workflow 35). Access requires Tailscale connection. |
How do I safely delete old artifacts in automation-hub logs? | Active generated artifacts are in /home/jade/automation-hub/data/logs/. Archived artifacts go to /home/jade/automation-hub/data/logs/archive/. Do not mix these with canonical docs or secrets. Archive before deleting — use the archive-review workflow (34) or manual copy before rm. Do not delete artifacts that are still ... |
What should I check before making any change to GlassHost backend code? | Read the file first to understand the existing pattern. Check AI module boundaries in /srv/glasshost/docs/AI_DIRECTION.md. Keep diffs minimal and reviewable. After changes: run bash /srv/glasshost/backend/tests/smoke_test.sh and ~/scripts/test-redaction.sh. If you changed runner_api.py in script-runner, rebuild and rec... |
When does the script-runner container need to be rebuilt? | Rebuild and recreate script-runner when changing /home/jade/automation-hub/scripts/runner/runner_api.py. Changes to scripts/guardrails/* usually only require a restart. Changes to scripts/ai_broker/* require restarting ai-broker. Workflow JSON changes require import into n8n. Doc changes require no runtime action. |
How do I import an n8n workflow after updating it? | Back up the current n8n state first: check /home/jade/automation-hub/docker/n8n_data/backups/ for a recent backup. Then import via the n8n UI at https://gg-core.egret-boa.ts.net:8444/ or via the n8n CLI if available. Do not CRUD workflows via raw n8n API from assistants — use the broker access path. |
What are the Rebuild/Restart/Import rules for this stack? | scripts/runner/runner_api.py changes → rebuild + recreate script-runner container
scripts/guardrails/* changes → restart relevant container
scripts/ai_broker/* changes → restart ai-broker
workflows/*.json changes → import into n8n
docs changes → no runtime action required |
How do I check if the automation hub is healthy? | From within the Docker network: curl http://automation-hub-script-runner:8080/health/dependencies
Or the automation status endpoint: GET /health/automation-status
The ops dashboard artifact is at /home/jade/automation-hub/data/logs/automation_ops_dashboard_last.json and .md. |
What is the AI broker handoff document? | The operator handoff is at /home/jade/automation-hub/docs/ai_broker_handoff.md. It documents the current state of automation, active workflows, recent incidents, and next operator steps. Review it at the start of any automation hub work session. |
How do I check the current automation ops dashboard? | Read /home/jade/automation-hub/data/logs/automation_ops_dashboard_last.md for the latest merged one-page view. For JSON: /home/jade/automation-hub/data/logs/automation_ops_dashboard_last.json. This is updated every 10 minutes by workflow 35. |
What is the Glass Government Reading Room? | The Reading Room is a planned feature for browsing ingested federal records. The implementation plan is at /home/jade/repos_clean/glassgovernment.org/docs/plans/2026-03-09-gg-reading-room-reader-impl.md and the design notes are at /home/jade/repos_clean/glassgovernment.org/docs/plans/2026-03-09-gg-reading-room-reader-d... |
What is the Glass Government Admin Hub? | The Admin Hub is a planned operator interface for managing the Glass Government stack. The implementation plan is at /home/jade/repos_clean/glassgovernment.org/docs/plans/2026-03-09-gg-admin-hub-impl.md and design notes at /home/jade/repos_clean/glassgovernment.org/docs/plans/2026-03-09-gg-admin-hub-design.md. |
How do I check the current backlog? | Read /home/jade/repos_clean/glassgovernment.org/docs/plans/2026-03-07-backlog-update-now-next-later.md for the Now/Next/Later breakdown. The backlog was refreshed toward the maintenance lane on 2026-03-12. |
What is the federal records integration audit? | The latest major audit is at /home/jade/docs/audits/2026-03-07-federal-records-integration-audit-and-plan.md. It covers ingestion pipeline status, federal source integrations (govinfo, muckrock, fbi_vault, NARA, EPA_ECHO), and the action plan for source coverage gaps. |
What does 'drift' mean in this stack? | Drift is a mismatch between documented expected state and actual live state. Examples: a port bound to 0.0.0.0 instead of documented 127.0.0.1, a workflow missing from automation_manifest.json but active in n8n, or a publisher path that has moved but docs still reference the old location. Drift causes operational confu... |
How do I verify there is no drift between the repo, staging, and production? | Use the parity workflow at /home/jade/repos_clean/glassgovernment.org/docs/ops/PARITY_WORKFLOW.md. Specifically run /home/jade/repos_clean/glassgovernment.org/scripts/snapshot-plugin-shas.sh and compare artifacts. For publisher ownership, check /srv/gg-python/current is the active path. |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- 1