desktop-agent-activity
Skill4 starsChanged 6 days agoMITView on GitHub
Use when 查 Orca/Claude Code 是否还在干活. Process+files first.
In this repository43 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- agent-symlink-initSkill
- code-inspector-initSkill
- commitSkill
- copywritingSkill
- css-tailwind-stylingSkill
- lingui-best-practicesSkill
- lingui-workflowSkill
- repo-skill-creatorSkill
- weekly-reportSkill
- agent-dispatchSkill
- agentic-review-handoffSkill
- agent-practice-auditSkill
- agent-symlink-initSkill
- architecture-hardening-loopSkill
- ardot-mcpSkill
- branch-creatorSkill
- chrome-dev-mcpSkill
- code-inspector-initSkill
- code-plugin-architectureSkill
- commit-pushSkill
- commitSkill
- decision-first-technical-writingSkill
- design-systemSkill
- dev-environment-roi-auditSkill
- disk-tidySkill
- figma-mcpSkill
- file-tidySkill
- goal-gateSkill
- grill-autopilotSkill
- kimi-computer-useSkill
- lingui-workflowSkill
- local-web-surfaceSkill
- open-code-review-loopSkill
- prompt-studioSkill
- review-prompt-composerSkill
- task-completion-loopSkill
- uxc-facadeSkill
- web-performance-auditSkill
- weekly-reportSkill
- workflow-gateSkill
- worktree-tidySkill
The file
Files
Desktop coding-agent activity
User asks whether a desktop-hosted coding agent (Orca + Claude Code, and similar PTY hosts) is still working, stuck, or idle.
Do not treat ps showing claude as "working". Alive sessions sit for hours with MCP children. Prove work with CPU + file/git mtimes + host telemetry.
Visual capture is optional confirmation, not the primary evidence.
Verdict (what to tell the user)
结论先说:在干 / 闲着挂着 / 已停。
表列每个实例:TTY 或 PID · 状态 · 依据(CPU 在跳 / 源文件刚改 / 终端停写多久)。≤5 行要点。不要贴命令输出。
Check order (do all of 1–4 before claiming idle)
- Process + cwd
psforclaude(and host app, e.g. Orca).lsof -a -p <pid> -d cwdonly — do not dump fulllsof -p(large Claude processes time out).- Parent is often
orca-tcc-login→ login zsh on attysXXX.
- CPU sample, 2–3 times over ~5s
- Working:
%CPUjumps (often 1–10%+). - Idle-alive: stays ~0, children are only MCP (
chrome-devtools-mcp,server-github,kimi-cu mcp). - Work-ish children:
caffeinate, compilers,git, test runners — not MCP.
- Working:
- TTY last-write + source mtimes
ls -lT /dev/ttysXXX— terminal still printing?- Walk the repo excluding
node_modules/.next/.gitfor.ts/.tsx/.mdmtime in the last hour. git status -sb+ recent commits. Uncommitted files with fresh mtimes = still editing.
- Host telemetry (Orca)
~/Library/Application Support/orca/agent-hooks/last-status.json— last prompt,state, worktree path.orca-stats.jsonagent_start/agent_stop.- Paths and parse pitfalls:
references/orca-claude-liveness.md.
Optional: screenshot
computer_use capture of the host window can confirm the TUI. If capture returns empty / 0×0, stop looping captures and ship the process+file verdict. Do not raise the window unless the user asked to look at the screen itself.
Pitfalls
- Alive ≠ working. MCP children spawn at session start and stay.
~/.claude/projects/**/*.jsonlmtime can lag days while the process is live. Never use stale jsonl as "not working".orca-statsunmatchedagent_startcan be months-old ghosts. Preferlast-status.json+ CPU + files.last-statusstate=donecan lag behind an already-resumed edit loop.state=workingcan stick on the last tool. Corroborate with file mtimes. Parse pitfalls:references/orca-claude-liveness.md.terminal-history/entries are directories, not files.- Iterating
Application Support/orcacan hit disappearing Electron cookies — skipFileNotFoundError, don't abort the check.
Out of scope
Starting, stopping, or steering the agent. This skill is read-only status.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

