Answer questions about this developer machine with real data from AGASTHY - repositories and their uncommitted work, installed runtimes, packages and tools, which process holds a port, which coding-agent sessions are running and what they left behind, token usage and estimated cost per agent, and what would break if a package were removed. Use when the user asks about their own machine, projects, ports, disk, packages, coding agents or token spend, or before suggesting they uninstall something.

AGASTHY: facts about this machine

agasthy is a read-only command-line tool. It never changes the machine: it reads what the AGASTHY desktop app has already scanned, and it reads live process and token-usage data. Every command prints JSON.

If agasthy is not installed

OpenClaw offers to download it. Elsewhere, tell the user to download the archive for their system from https://github.com/HMAKT99/Agasthy/releases/latest (agasthy-cli-macos-aarch64.tar.gz, agasthy-cli-linux-x86_64.tar.gz or agasthy-cli-windows-x86_64.zip), unpack it, and put agasthy on their PATH. Each archive has a .sha256 checksum next to it. Do not download or install it yourself unless they ask.

How to answer

  1. Start with agasthy status. It shows when each kind of data was last scanned (age_hours). Mention the age when it matters ("as of the scan 6 hours ago"). If it says there is no database, the desktop app has not run yet: suggest installing AGASTHY from https://github.com/HMAKT99/Agasthy/releases/latest and running a scan, and use the live commands below meanwhile.
  2. Pick the narrowest command from the table. Prefer queries over rescans; they are instant.
  3. Answer from the JSON only. Quote names, paths, versions and numbers exactly. A value with "status": "unavailable" means unknown, never zero; say it is unknown.
  4. Never act on the machine for the user (kill processes, uninstall, delete). Explain what AGASTHY found and point them to the AGASTHY app, which reviews every change with a dry run and lets them undo it. Suggest commands for the user to run only when they ask.

Commands

QuestionCommand
How fresh is the data?agasthy status
Find a repository, package, runtime, tool, app, MCP server...agasthy find <text> [--kind KIND] [--limit N] (add --full for all attributes)
Everything about one item, with evidence and relationshipsagasthy entity <id> (ids come from find)
What breaks if I remove X?agasthy impact <id> (gives risk, affected projects, a plan and how to recover)
Which dependencies do most projects share?agasthy shared
Which coding agents are running, and what did they leave running?agasthy agents
What is using port N / which process belongs to which project?agasthy processes
How many tokens did my agents use, and what did it cost?agasthy usage [--days N] (totals per agent; add --full for sessions and days, each with agent, cwd (the project folder), tokens and estimated_cost_usd, plus models per session or model and day per day)
Memory, CPU, disk right nowagasthy telemetry

KIND is one of: repository, application, binary, runtime, package, package_manager, tool, ai_tool, ai_provider, mcp_server, ai_config_file, finding, remote_project.

Reading the results

  • Uncommitted work: in agasthy find <name> --kind repository --full, see attributes.working_tree.value (modified, untracked, staged, clean) and attributes.head.value.branch.
  • Agents: each item in sessions has label (Claude Code, Codex, Cursor, Gemini CLI, Hermes, OpenClaw, ...), project, total_memory_bytes, ports, mcp_servers and dev_servers. leftovers are processes still running after the agent that started them exited; agent_label says which agent, when known.
  • Ports: in agasthy processes, each item in processes has its own ports (the ports it listens on), plus name, command, cwd, memory_bytes and the project it belongs to. The top-level ports is only how many listening ports were found.
  • Token cost is an estimate from public API list prices. People on a subscription plan are billed differently; say so when you quote a cost. Usage reads only token counters from each agent's local logs, never message text.
  • Impact: risk is low, medium or high. runtime_reports[].items[].outcome says, per project, whether it breaks or falls_back to another installed copy.

Privacy

Results contain local paths, project names and process command lines. Use them to answer the user here. Do not send them to other services or post them anywhere unless the user asks.