Data Layer

Data Layer - cross-harness monitoring for the portable brain

Use this skill when the user wants to measure agent activity across Claude Code, Hermes, OpenClaw, Codex, Cursor, OpenCode, Windsurf, Pi, Antigravity, or any custom loop using .agent/.

The goal is local business intelligence for the whole agent suite:

• what harnesses are active
• how many agent events are happening
• when cron/scheduled agents fire
• which crons started/finished and how long they ran
• how many agents are active
• tokens and estimated cost by hour/day/week/month
• resource usage by user-defined category
• workflow success/error rates
• KPI summary rows for cron cadence, run volume, reliability, active agents, workflow breadth, token usage, and estimated cost
• terminal dashboard visible directly in the user's coding tool
• screenshot-ready daily resource reports

Hard Rules

• Stay local-first. Do not add telemetry or remote sync.
• Do not store raw prompts, raw code, client names, emails, phone numbers, or unredacted business records in shared examples.
• Do not commit .agent/data-layer/ exports unless the user explicitly reviewed and sanitized them.
• Do not send dashboard screenshots to email, Slack, webhooks, or any other channel unless the user explicitly approves the destination.

Inputs

Default inputs:

.agent/memory/episodic/AGENT_LEARNINGS.jsonl
.agent/data-layer/harness-events.jsonl     optional
.agent/data-layer/cron-runs.jsonl          optional
.agent/data-layer/category-rules.json      optional

AGENT_LEARNINGS.jsonl is the shared activity log. Optional files let users add events from harnesses that do not automatically write rich events yet.

Agent behavior

When this skill is injected, decide whether the user is asking to see local agent activity. Natural prompts such as "what did my agents do", "show me the dashboard", "how many tokens did we use", or "show last week by hour" should render the terminal dashboard directly. Do not make users remember flags.

Prefer passing the user's words to the exporter:

python3 .agent/tools/data_layer_export.py show me last 7 days by hour

If the user gives no range or bucket, run the default export. Explicit flags still work for scripts and should override the natural-language words.

Export

Run:

python3 .agent/tools/data_layer_export.py --window 30d --bucket day

The command prints a compact terminal dashboard by default, then writes the full browser dashboard and data files.

Use --bucket hour, --bucket day, --bucket week, or --bucket month for different chart grains.

Outputs go to:

.agent/data-layer/exports/<YYYY-MM-DD>/

Key outputs:

• agent-events.jsonl/csv
• cron-runs.jsonl/csv
• cron-timeline.json/csv
• activity-series.json/csv
• category-summary.json/csv
• harness-summary.json/csv
• workflow-summary.json/csv
• kpi-summary.json/csv
• dashboard-summary.json
• dashboard-report.json
• dashboard.html
• dashboard.tui.txt
• daily-report.md

Categories

Users can define any categories they want in .agent/data-layer/category-rules.json, for example:

{
  "default_category": "uncategorized",
  "rules": [
    {"category": "coding", "skills": ["debug-investigator", "git-proxy"]},
    {"category": "admin", "run_types": ["cron"]},
    {"category": "financial", "workflows": ["invoice_collection"]},
    {"category": "personal", "workflows": ["calendar_coordination"]},
    {"category": "work", "phases": ["plan", "review", "qa", "ship"]}
  ]
}

Daily Screenshot Report

For daily resource management:

1. Run the exporter from a user-approved cron or scheduled agent.
2. Open dashboard.html.
3. Capture the Resource Overview, Activity, Tokens, Cron Runs, Task Categories, Harness Mix, Workflow Outcomes, Cron Gantt, and Cron Timeline sections.
4. Send the screenshot only through an explicitly approved channel.

The exporter creates daily-report.md as the text summary. It does not send anything by itself.

Self-rewrite hook

If exports are confusing twice in a row, improve category rules, schema names, or dashboard labels before adding heavier dashboard dependencies.