xagent
Skill4 starsChanged 5 months agoMITView on GitHub
Dispatch a single instruction to another coding-agent CLI such as Codex, Claude, or Cursor Agent when the user wants an external review, explanation, or sanity check while keeping the external trace out of the current context.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- xiaoyuge886/aigctheir own version, first seen Oct 8, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- xiaoyuge886/aigc★ 198same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- xflowSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 0 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 30
Prefer the explicit form so word order and shell characters can't break it — pass the task as one quoted --instruction:
- Very specific: line 37
Do not create run directories, choose output paths, or compose prompt files.
- Very specific: line 40
Read stdout.log / stderr.log (prefer tail) only when explicitly debugging a failure.
- Very specific: line 55
Default resolution: $XAGENT_REGISTRY when set, else ~/.claude/xagent-registry.json if it exists, else the skill's bundled …
- Very specific: line 75
Always exits 0; status is in stdout and meta.json.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
xagent
Dispatch one task to one external coding agent. The agent's full trace stays in log files under .xagent-runs/; you see only a bounded structured summary.
Security
xagent dispatches autonomous CLIs in non-interactive mode. Use it only for trusted instructions and trusted repositories.
Commands
/xagent
/xagent list
/xagent validate
/xagent codex review the diff
/xagent claude explain this function
/xagent cursor sanity-check this approach
/xagent --show codex review the diff
/xagent --model sonnet claude review the diffBehavior
- Invoke the runner with
python3by its absolute path —<skill>/scripts/xagent_run.py, where<skill>is this skill's directory — and run it from the user's current project directory, not from inside the skill directory. Omit--project-rootby default: the runner auto-detects the project and echoes it asproject_root:in the summary. If the echoed root is wrong, or the runner reports it was launched from inside the skill directory (reason: config_error), re-run with--project-root <repo>. - Prefer the explicit form so word order and shell characters can't break it — pass the task as one quoted
--instruction:
The shorthandpython3 <skill>/scripts/xagent_run.py --agent <name> --instruction "<the user's task>"… <name> <task words…>also works, but any runner flags (--show,--max-time, …) must come before<name>. Reference files in the instruction by repo-root-relative or absolute paths: the dispatched agent's working directory is the git top-level, not necessarily the user's subdirectory. - For
/xagentor/xagent list, runpython3 <skill>/scripts/xagent_run.py listto print available agents, then stop. - For
/xagent validate, runpython3 <skill>/scripts/xagent_run.py validate --update-tested-version(this flag implies--smoke). This validates every active registry agent in one temporary project, checks file creation with a nonce, and pinstested_versionfor each agent that passes — failing agents are skipped and reported, not blocking the rest. - Do not create run directories, choose output paths, or compose prompt files. The runner resolves the registry, run directory,
prompt.md, andoutput.mditself. - Show the runner's structured summary to the user verbatim; do not paste log contents.
- For follow-up questions, read the file named in the summary's
output_file:line — not the stdout/stderr logs. - Read
stdout.log/stderr.log(prefertail) only when explicitly debugging a failure.
Result
The summary's status: field is one of:
success—output.mdwas produced; the answer is inoutput_file.partial— output exists but the agent exited nonzero or was timed out; usable but incomplete.timeout— killed with no output.error— no usable output;reasonsays why (empty_output,unparseable_output,child_failed,child_fatal,unknown_agent,config_error,runner_error). Onchild_failed/child_fatalthe summary carries adiagnostic:block with the stderr tail.child_fatalmeans the runner detected a non-recoverable stderr signal (backend quota/capacity exhaustion or an auth failure) and aborted the agent early rather than waiting out the timeout; thediagnostic:block leads with the matchedfatal signal:line. Do not blindly re-dispatch achild_fatalrun — fix the underlying condition (re-auth, wait for quota, or switch model/agent) first.
The summary also carries project_root: and run_dir: lines — check that project_root: matches the user's project before trusting the result. On config_error, fix the invocation or registry. On runner_error, it is an internal failure — report it. A version_warning: line means the agent's installed version differs from its recorded tested_version. The runner always exits 0, so branch on status / reason, never the exit code.
Registry
Default resolution: $XAGENT_REGISTRY when set, else ~/.claude/xagent-registry.json if it exists, else the skill's bundled assets/xagent-registry.json.
An entry with "active": false is disabled: hidden from list, refused on dispatch (reason=inactive_agent), and skipped by validate.
See references/registry-format.md for schema details.
The runner uses stdlib json only — no pip dependencies required for dispatch.
Runner
python3 <skill>/scripts/xagent_run.py --agent <name> --instruction "<task>"
python3 <skill>/scripts/xagent_run.py validate --smoke --update-tested-version- The runner creates the run directory and renders the prompt itself; it writes
.xagent-runs/.gitignore(*) automatically so traces are never committed. - Pruning keeps the newest 20 terminal runs (dirs with
meta.json); dirs withoutmeta.jsonare removed once older than 7 days. - Defaults:
--project-root= git top-level of the working directory (else the working directory);--registry=$XAGENT_REGISTRY, else~/.claude/xagent-registry.jsonif it exists, else the skill's bundledassets/xagent-registry.json;--output-file=<run-dir>/output.md. - Overrides (rarely needed):
--run-dir <abs>pins a run location;--project-root <repo>targets a different repo;--prompt-file <abs>supplies a pre-rendered prompt instead of--instruction;--model <name>overrides the registry's default model for this dispatch. - Liveness resets on streamed stdout/stderr, output-file growth, process CPU, and the active Codex rollout log.
--max-timeis an idle backstop, so active runs may exceed it; the inactivity timeout remains the normal hung-child limit. --showprints the resolved argv and prompt without spawning the agent.validatechecks every active registry entry (agents withactive: falseare skipped). Without flags it performs static registry, argv, binary, and version-command checks. With--smoke, it dispatches every agent in one temporary project and verifies output plus a nonce-bearing sentinel file.--update-tested-versionimplies--smokeand pinstested_versionper agent: each agent that passes is updated, each that fails (or has no parseable version) is skipped and listed underskipped (version pin):.- Always exits 0; status is in stdout and
meta.json.
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.

