Agent Pulse Skills
Codex skill for using Agent Pulse to inspect local AI-agent sessions, token usage, model costs, health, reports, and MCP tools.
This repository is a skill package. The actual Agent operating guide lives in agent-pulse/SKILL.md.
Install
Install from GitHub with the skills CLI:
npx skills add Jane-o-O-o-O/agent-pulse-skills
List the skills exposed by this repository:
npx skills add Jane-o-O-o-O/agent-pulse-skills --list
Manual local install:
git clone git@github.com:Jane-o-O-o-O/agent-pulse-skills.git
cd agent-pulse-skills
npx skills add .
Runtime Requirement
The skill tells an agent how to use Agent Pulse, but the agent-pulse CLI must be installed separately:
pip install agentpulse-cli
agent-pulse --version
For local Agent Pulse development:
cd path/to/agent-pulse
pip install -e ".[web]"
agent-pulse --version
What Agents Use It For
| User asks about | Agent Pulse commands the skill points to |
|---|---|
| Recent usage | status, top, root dashboard JSON |
| Sessions and search | session, search, diff, snapshot |
| Models and cost | models, leaderboard, optimize, budget |
| Risk and health | alerts, health, anomaly, forecast, score |
| Trends and comparisons | history, timeline, heatmap, compare, compare-projects, insights |
| Sharing and integration | report, export, export-html, metrics, web, api, mcp |
| Setup and discovery | doctor, scan, config, plugins, frameworks, completions |
Supported source keys for -P/--platform:
hermes, claude, codex, deepseek, openclaw, copilot, aider, qwen,
opencode, goose, cursor, antigravity, amp
Example Prompts
Use $agent-pulse to summarize my AI agent usage in the last 24 hours.
Use $agent-pulse to check which model cost the most this week.
Use $agent-pulse to diagnose why my dashboard has no sessions.
Use $agent-pulse to compare Codex and Claude activity over the last 7 days.
Use $agent-pulse to list the MCP tools Agent Pulse exposes.
Included Files
agent-pulse/
|-- SKILL.md
|-- agents/
| `-- openai.yaml
`-- scripts/
`-- run_agent_pulse_snapshot.py
The helper script runs a compact JSON-friendly snapshot:
python agent-pulse/scripts/run_agent_pulse_snapshot.py --hours 24 --days 7
It gathers diagnosis, current status, top cost sessions, model analytics, leaderboard, forecast, health, score, budget, and insights. Each subcommand has its own timeout so one slow local scan does not block the whole snapshot.
Windows Notes
Agent Pulse prints emoji and box-drawing characters. Use UTF-8 output for reliable command execution:
$env:PYTHONUTF8='1'
$env:PYTHONIOENCODING='utf-8'
Validation
Run the skill validator from the skill-creator skill:
python path/to/skill-creator/scripts/quick_validate.py path/to/agent-pulse
Expected result:
Skill is valid!