How do I install the Tf Maintenance skill?

Run npx skillmds@latest add monoscope-tech/tf-maintenance in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Tf Maintenance safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Tf Maintenance?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Tf Maintenance free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Tf Maintenance?

monoscope-tech (@monoscope-tech) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Tf Maintenance by monoscope-tech — Agent Skill

Name: Tf Maintenance
Author: monoscope-tech

Tf Maintenance

Refresh and republish the intra-day maintenance dashboard (ingest/flush, hot-tail compaction, dedup queue, crons, read path, OOM banner) at its permanent artifact URL. Use for "what's the server doing today", light-compaction/dedup status, or "update the maintenance page".

File contents

Refresh docs/dashboards/hot-maintenance.html with live data and republish to the permanent artifact URL below (pass it as url to the Artifact tool from any session — never mint a new URL).

Artifact URL (stable): https://claude.ai/code/artifact/b3b569fc-56e8-4527-8600-b3e5e80774ce · favicon ⚙️

1. Data sources

Counters (reset at each boot — always pair numbers with boot time):

PGURL=$(grep -m1 '^TIMEFUSION_PG_URL=' ../monoscope/.env | cut -d= -f2-)
psql "$PGURL" -t -A -F'|' -c "SELECT component,key,value FROM timefusion_stats"

Key rows: buffered_layer (rows_ingested_total, flush_completed_total, pressure_pct, process_rss_mb), maintenance (light_optimize_* waves/bins/ticks, dirty_bin_queue_depth, dedup_*, checkpoints_created), wal (quarantine_files/mb, disk_mb), pgwire/scan/plan_cache/foyer for the read path.

Restart/OOM state (read-only ssh):

ssh ubuntu@captain.s.past3.tech 'docker service ps srv-captain--timefusion --format "{{.Name}} {{.CurrentState}} {{.Error}}" | head -3'
# OOM confirmation: docker inspect <exited container> --format "OOM:{{.State.OOMKilled}} at {{.State.FinishedAt}}"

pgwire.queries_total doubling back to ~0 also betrays a restart.

2. Update the template

Edit docs/dashboards/hot-maintenance.html: the red .banner (OOM/restart story — keep the history, e.g. "kills at 18:47 and 20:32"), every .tile number, the .sub boot-time line, and the closing .note. Health pills: ok/warn/crit.

Interpretation guide:

MemBuffer pressure low + RSS tens of GB = heap churn / reservation-leak pattern (see memory tf_compaction_binfanin_leak_2026-07-30, OOM #3).
dirty_bin_queue_depth in the thousands with dedup_bins_deferred_cold close behind is by-design cold backlog; read-side DedupExec keeps queries correct regardless.
WAL quarantine_files > 0 = acked data not in the store → investigate (quarantine/dml/* = parked enrichment merges; timefusion redrive-dml recovers them — see memory tf_dml_quarantine_redrive_2026-07-30 if present, else the 800373f commit message).

3. Republish

Artifact tool with file_path: docs/dashboards/hot-maintenance.html, url: <the stable URL above>, favicon ⚙️.

---name: tf-maintenancedescription: Refresh and republish the intra-day maintenance dashboard (ingest/flush, hot-tail compaction, dedup queue, crons, read path, OOM banner) at its permanent artifact URL. Use for "what's the server doing today", light-compaction/dedup status, or "update the maintenance page".---Refresh `docs/dashboards/hot-maintenance.html` with live data and republish tothe permanent artifact URL below (pass it as `url` to the Artifact tool fromany session — never mint a new URL).**Artifact URL (stable):** https://claude.ai/code/artifact/b3b569fc-56e8-4527-8600-b3e5e80774ce · favicon `⚙️`## 1. Data sources**Counters** (reset at each boot — always pair numbers with boot time):```bashPGURL=$(grep -m1 '^TIMEFUSION_PG_URL=' ../monoscope/.env | cut -d= -f2-)psql "$PGURL" -t -A -F'|' -c "SELECT component,key,value FROM timefusion_stats"```Key rows: buffered_layer (rows_ingested_total, flush_completed_total,pressure_pct, process_rss_mb), maintenance (light_optimize_* waves/bins/ticks,dirty_bin_queue_depth, dedup_*, checkpoints_created), wal (quarantine_files/mb,disk_mb), pgwire/scan/plan_cache/foyer for the read path.**Restart/OOM state** (read-only ssh):```bashssh ubuntu@captain.s.past3.tech 'docker service ps srv-captain--timefusion --format "{{.Name}} {{.CurrentState}} {{.Error}}" | head -3'# OOM confirmation: docker inspect <exited container> --format "OOM:{{.State.OOMKilled}} at {{.State.FinishedAt}}"````pgwire.queries_total` doubling back to ~0 also betrays a restart.## 2. Update the templateEdit `docs/dashboards/hot-maintenance.html`: the red `.banner` (OOM/restartstory — keep the history, e.g. "kills at 18:47 and 20:32"), every `.tile`number, the `.sub` boot-time line, and the closing `.note`. Health pills:`ok`/`warn`/`crit`.Interpretation guide:- MemBuffer pressure low + RSS tens of GB = heap churn / reservation-leak pattern (see memory `tf_compaction_binfanin_leak_2026-07-30`, OOM #3).- `dirty_bin_queue_depth` in the thousands with `dedup_bins_deferred_cold` close behind is by-design cold backlog; read-side DedupExec keeps queries correct regardless.- WAL `quarantine_files > 0` = acked data not in the store → investigate (`quarantine/dml/*` = parked enrichment merges; `timefusion redrive-dml` recovers them — see memory `tf_dml_quarantine_redrive_2026-07-30` if present, else the 800373f commit message).## 3. RepublishArtifact tool with `file_path: docs/dashboards/hot-maintenance.html`,`url: <the stable URL above>`, favicon `⚙️`.

Frequently asked questions