$ ~/registry/skill/openclaw-coding-agent

SKILL

coding-agent

Name: coding-agent
Author: openclaw

Delegate coding tasks to background AI coding agents for implementation and changes.

Stars

★ 382,582

Source

GitHub

Updated

2026-07-11

// securityLow risk

Prompt only, no code
Open source
Community-trusted· 377.5k

Summary

The material indicates this is essentially an open-source prompt-only orchestration guide for invoking external coding agents; it does not itself declare secrets, remote endpoints, or built-in execution logic. Overall risk is low, though the instructions do steer the host toward launching background coding agents and local commands, so execution boundaries on the host side still matter.

CredentialsLow risk

The material explicitly states there are no required secrets or environment variables, and the README does not ask for API tokens or long-lived credentials. No direct credential collection, storage, or exfiltration behavior is described. If the host later invokes Codex/Claude Code/OpenCode, their credential handling belongs to those external tools rather than this skill itself.

Network egressLow risk

No remote endpoints are declared, and the material does not specify any fixed exfiltration host or third-party receiver. The README mentions GitHub issues/PRs, notification routes, and message sending, but these appear to be workflow instructions for external systems; this skill itself does not define built-in network destinations.

Code executionLow risk

The objective check marks it as prompt-only, meaning the skill itself is more of an instruction set than an executable component. While the README heavily describes launching Codex/Claude/OpenCode workers via bash, PTY, and background processes, those are host/external-agent usage instructions and do not mean the skill itself inherently possesses local execution privileges.

Data accessLow risk

The material does not declare any direct ability for this skill to read or write local files, repositories, or system resources, nor does it request extra directory or device permissions. The README does suggest running external workers in repo directories, temp files, and isolated checkouts, but that is workflow guidance for external execution; no overbroad access is evident for the skill itself.

Supply chainLow risk

The source is an open-source GitHub repository with extremely high community adoption (about 377k stars), both of which are strong risk-reducing signals. The material also shows no closed-source binary fetches, suspicious install scripts, or unclear distribution channels. The missing license declaration and unknown maintenance status are worth verifying, but based on current evidence the supply-chain risk remains low.

Recommendations

Treat this skill as a prompt/orchestration template and do not automatically grant high-privilege execution environments to the external agents it references.
If you implement the background-worker workflow from the README, prefer isolated repos, least-privilege accounts, and controlled notification channels.
Verify the repository license, recent commits, and issue handling to confirm the project is still actively maintained.
Separate the risk boundary of this skill from the external coding agents it invokes, and audit Codex, Claude Code, and OpenCode individually for credentials, network, and execution permissions.

Audited by: gpt-5.4 · 2026-06-16

// install

Copy the install command and let the AI configure it · recommended for beginners

Please install the "coding-agent" skill from askskill:
1. Download https://raw.githubusercontent.com/openclaw/openclaw/main/skills/coding-agent/SKILL.md
2. Save it as ~/.claude/skills/coding-agent/SKILL.md
3. Reload skills and tell me it's ready

// download

Download SKILL.mdInstall manifest ↗

// examples

Implement a new feature in batch

Prompt

Delegate the "export user profiles to CSV" feature to the background coding-agent. The stack is Node.js + Express. Add an export endpoint, field mapping, permission checks, and unit tests, then summarize changed files and how to run it.

Expected

A completed implementation summary with code changes, affected files, test status, and suggested next steps.

Refactor a legacy module

Prompt

Delegate the refactoring of this Python payment module to coding-agent. Split long functions, add type hints, remove duplicated logic, and keep existing API behavior unchanged. Provide a refactoring summary, risks, and recommended validation steps.

Expected

Receive cleaner refactored code with compatibility notes and recommended validation steps.

Fix a backend error and add tests

Prompt

Assign this Django API 500-error fix to coding-agent. Use the error logs to locate the issue, implement the fix, and add regression tests with a brief change summary. Don’t just analyze—complete the code modification plan.

Expected

An actionable fix result describing the root cause, solution, test coverage, and possible impact scope.

// docs

Coding Agent

Use for background feature builds, PR reviews, large refactors, and issue-to-PR loops. Do not use for simple edits, read-only lookup, ACP thread-bound work, or any run inside ~/.openclaw, $OPENCLAW_STATE_DIR, or active OpenClaw state dirs.

Hard rules

Always launch with background:true.
Codex and OpenCode: use pty:true.
Claude Code: no PTY; use claude --permission-mode bypassPermissions --print.
Capture a real notification route before spawning.
Worker must send completion/failure via openclaw message send.
Do not rely on heartbeat, system events, or notify-on-exit.
Monitor with process; do not kill slow workers without cause.
If user asked for a specific agent, use that agent.
If worker fails/hangs, respawn or ask; do not silently hand-code instead.
Never checkout branches or run background coding agents in ~/Projects/openclaw; use an isolated checkout.

Notification block

Append this shape to every worker prompt with real values:

Notification route:
- channel: <notifyChannel>
- target: <notifyTarget>
- account: <notifyAccount or omit>
- reply_to: <notifyReplyTo or omit>
- thread_id: <notifyThreadId or omit>

When finished, send exactly one completion or failure message using:
openclaw message send --channel <channel> --target '<target>' --message '<brief result>'
Add --account, --reply-to, or --thread-id only when present above.
Do not use openclaw system event or heartbeat.

If no trustworthy route exists, say completion auto-notify is unavailable.

Launch forms

Write the worker prompt to a temp file first. This avoids shell quoting bugs when the required notification block contains quotes or newlines.

PROMPT=$(mktemp -t openclaw-worker-prompt.XXXXXX)
cat >"$PROMPT" <<'EOF'
Task.
<notification block>
EOF
printf 'prompt file: %s\n' "$PROMPT"

Use $PROMPT when launching from the same shell/session. If using a separate tool call, substitute the printed path.

Codex:

bash pty:true background:true workdir:/path/repo command:"codex exec - < \"$PROMPT\""

Claude Code:

bash background:true workdir:/path/repo command:"claude --permission-mode bypassPermissions --print < \"$PROMPT\""

OpenCode:

bash pty:true background:true workdir:/path/repo command:"opencode run < \"$PROMPT\""

Long issue-to-PR work

Create/reuse a GitHub issue as durable spec.
Include issue URL, repo, base branch, expected PR, proof, and notification route.
Tell worker to branch, implement, test, run review until no accepted actionable findings, open PR.
Return issue URL and sessionId immediately.
Monitor with process; cancel through Task Registry if mirrored there.

Scratch Codex

Codex needs a trusted git repo:

SCRATCH=$(mktemp -d)
git -C "$SCRATCH" init
PROMPT=$(mktemp -t openclaw-worker-prompt.XXXXXX)
cat >"$PROMPT" <<'EOF'
Build X.
<notification block>
EOF
printf 'prompt file: %s\n' "$PROMPT"
bash pty:true background:true workdir:$SCRATCH command:"codex exec - < \"$PROMPT\""

Process actions

list: running/recent sessions.
poll: status.
log: output.
submit: send input + Enter.
write: raw stdin.
paste: paste text.
kill: terminate.

Status to user

Say what started, where, and sessionId.
Update only on milestone, worker question, error, user action needed, or finish.
If killed, say why.

MCP

cc-in-codex

Coordinate Codex and local Claude Code for engineering, review, and automation workflows.

—Get→

MCP

gpt-subagents-api

Delegate coding tasks to OpenAI subagents with parallel, cheap, verified orchestration.

—Get→

$ loading_