Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/arc0btc-agent-health
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Node.js package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/agent-health
What's inside
- Skill name
agent-health- Description
- External health monitor for Loom (Rising Leviathan). Analyzes pre-gathered cycle metrics, task failures, git activity, and gate state to classify agent health and send email alerts.
- License
- MIT
- Instructions
- 139 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (3)
- SKILL.md
- loom.json · agents/
- sensor.ts
SKILL.md (first lines)
# Agent Health Monitor
## Purpose
This skill is loaded by Haiku tasks created by the `agent-health` sensor running on Arc (Trustless Indra). Arc SSHes into Loom every 2 hours, gathers all relevant data, and bakes it into the task description as a structured data block.
**Your only job as Haiku:** read the pre-gathered data, classify health status (GREEN / YELLOW / RED), and send an email if anything needs attention. You do not need to SSH anywhere or query any databases -- everything is already in the task description.
## Data Block Format
The task description will contain a block structured like this:
```
=== AGENT HEALTH DATA: Loom (Rising Leviathan) ===
Checked: {ISO timestamp}
Period: last {N} hours
## Cycle Metrics
cycles_checked: N
total_tokens_in: N
total_cost_usd: N.NN
avg_tokens_per_cycle: N
avg_cost_per_cycle: N.NN
max_tokens_in_single_cycle: N
max_cost_single_cycle: N.NN
spike_cycles: [{started_at, tokens_in, cost_usd, model, duration_ms, task_subject}]
## Task Failures
failed_sources: [{source, count, last_subject}]
retry_storms: [{source, count, subjects}]
## Git Activity
commits_since_last_check: N
watched_path_commits: [{hash, date, message, files_changed}]
## Gate State
dispatch_gate: {status, consecutive_failures, stopped_at, stop_reason}
watchdog: {last_ran, last_alert_at, last_result}
## Thresholds
tokens_in_per_cycle: 1000000
cost_per_cycle_usd: 3.00
daily_cost_usd: 100
failed_source_repeat_count: 5
pending_task_age_hours: 4
=== END DATA ===
```
## Analysis Instructions
Work through these steps in order:
### Step 1: Compare Each Metric Against Its Threshold
Check each of the following, noting which are within bounds and which exceed thresholds:
| Metric | Threshold | Signal |
|--------|-----------|--------|
| max_tokens_in_single_cycle | 1,000,000 | RED if exceeded |
| max_cost_single_cycle | $3.00 | RED if exceeded |
| total_cost_usd (for period) | $100/day equivalent | YELLOW if on pace to exceed |
| failed_source repeat count | 5 occurrences | RED if any source exceeds this |
| watched_path git commits | any non-trivial changes | YELLOW if behavior-critical code changed |
| pending task age | 4 hours | YELLOW if any task pending longer |
| dispatch_gate status | "open" | RED if stopped or closed |
### Step 2: Apply Known Noise Filters
Skip the following when counting failures (these sensors create/fail tasks as normal operation):
- `sensor:treasury-health`
- `sensor:service-health`
- `sensor:service-health:stale-lock`
Do not flag these sources as retry storms even if their counts are high.
### Step 3: Classify Overall Status
- **GREEN**: All metrics within thresholds, no watched-path commits, gate open
- **YELLOW**: At least one metric in warning range, OR non-trivial watched-path commits, OR any pending task past age threshold
- **RED**: Any threshold critically exceeded, OR dispatch gate stopped, OR retry storm detected from a non-noise source
If multiple signals exist, use the worst classification.
### Step 4: Decide Whether to Send Email
- **GREEN**: Do not send email. Mark the task complete with a one-line "all clear" note.
- **YELLOW**: Send email to whoabuddy@gmail.com with subject `[Loom Health] YELLOW - {brief reason}`.
- **RED**: Send email to whoabuddy@gmail.com with subject `[Loom Health] RED - {brief reason}`.
### Step 5: Compose Email Body
For YELLOW emails, include:
- Overall status classification
- Which metrics triggered the warning
- Relevant data points (not the full dump, just what's relevant)
- Suggested investigation steps
For RED emails, include:
- Overall status classification (urgent)
- Which thresholds were exceeded and by how much
- Full data for the flagged metrics (paste the relevant sections from the data block)
- Immediate action items (e.g., "check if Loom is running a token spiral", "inspect recent git commits to src/")
### Step 6: Send Email Using arc-email-sync
Use the `arc-email-sync` skill to send the email:
- To: whoabuddy@gmail.com
- From: arc@arc0.me (default)
- Subject: as defined above
- Body: as composed above
## Email Quick Reference
| Status | Action | Subject Pattern |
|--------|--------|-----------------|
| GREEN | No email, note "all clear" | (none) |
| YELLOW | Send warning email | `[Loom Health] YELLOW - {reason}` |
| RED | Send urgent email | `[Loom Health] RED - {reason}` |From arc0btc/arc-starter, MIT.
More from arc0btc/arc-starter
- 0.544score
aibtc-news-editor
biwasxyz
Beat Editor for aibtc.news: review and approve/reject signals on assigned beat, file editorial reviews, manage beat cap via displacement, earn per-review sats
SkillBFreshness: B★ 1Claude Code +9 - 0.544score
arc-article-pipeline
arc0btc
Select a crown-jewel research finding, draft a long-form arc0.me article + an X Article amplification variant (title + body + companion post, emailed…
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
arc-architecture-review
arc0btc
Continuous architecture review, state machine diagrams, and simplification via SpaceX 5-step process
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-news-editorial
arc0btc
File intelligence signals, claim editorial beats, track correspondent activity on aibtc.news
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-dev-ops
arc0btc
Monitor service health via worker-logs and enforce production-grade standards across all aibtcdev repos
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-repo-maintenance
arc0btc
Triage, review, test, and support aibtcdev repos we depend on
SkillBFreshness: B★ 1Claude Code +9 - 0.494score
arc-artifacts
arc0btc
Vacuum + audit CLI for the source-artifact pool (artifacts/distilled/ + distilled_artifacts table).
SkillCFreshness: C★ 1Claude Code +9 - 0.494score
agent-pitch
arc0btc
Framing and language for pitching the AIBTC / agent-economy thesis to external audiences — talks, blog posts, decks, threads
SkillCFreshness: C★ 1Claude Code +9 - 0.494score
aibtc-news-classifieds
arc0btc
Classified ads and extended API coverage for aibtc.news — list, post, and manage classifieds; read briefs; correct signals; update beats; fetch streaks and…
SkillCFreshness: C★ 1Claude Code +9 - 0.494score
arc-attribution
arc0btc
MRR / attribution report reconciling Whop + x402 sales, channel attribution, and reach into one source of truth for demand-flywheel measurement
SkillCFreshness: C★ 1Claude Code +9 - 0.491score
arc-blocked-review
arc0btc
Sensor that periodically reviews blocked tasks to check if they can be unblocked
SkillCFreshness: C★ 1Claude Code +9 - 0.491score
aibtc-agent-trading
arc0btc
Detects AIBTC-network agent trading activity from JingSwap cycles and P2P ordinals desk for agent-trading beat signals
SkillCFreshness: C★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 4
- Open issues
- 3
- Last commit (this folder)
- Jul 14, 2026 (3 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/arc0btc/arc-starter
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from arc0btc
- 0.544score
arc-article-pipeline
arc0btc
Select a crown-jewel research finding, draft a long-form arc0.me article + an X Article amplification variant (title + body + companion post, emailed…
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-repo-maintenance
arc0btc
Triage, review, test, and support aibtcdev repos we depend on
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-dev-ops
arc0btc
Monitor service health via worker-logs and enforce production-grade standards across all aibtcdev repos
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
arc-architecture-review
arc0btc
Continuous architecture review, state machine diagrams, and simplification via SpaceX 5-step process
SkillBFreshness: B★ 1Claude Code +9 - 0.541score
aibtc-news-editorial
arc0btc
File intelligence signals, claim editorial beats, track correspondent activity on aibtc.news
SkillBFreshness: B★ 1Claude Code +9 - 0.494score
agent-pitch
arc0btc
Framing and language for pitching the AIBTC / agent-economy thesis to external audiences — talks, blog posts, decks, threads
SkillCFreshness: C★ 1Claude Code +9
Cite this page
agent-health skill. Agent Mart. https://agentmart.dev/skills/arc0btc-agent-health (last updated 2026-10-24)