coding-agent
Implementation/review workflow for approved plans. Route ACP-first, then CLI fallback. Use after explicit APPROVE.
Best use case
coding-agent is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Implementation/review workflow for approved plans. Route ACP-first, then CLI fallback. Use after explicit APPROVE.
Teams using coding-agent should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/coding-agent/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How coding-agent Compares
| Feature / Agent | coding-agent | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Implementation/review workflow for approved plans. Route ACP-first, then CLI fallback. Use after explicit APPROVE.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
AI Agents for Coding
Browse AI agent skills for coding, debugging, testing, refactoring, code review, and developer workflows across Claude, Cursor, and Codex.
Best AI Skills for Claude
Explore the best AI skills for Claude and Claude Code across coding, research, workflow automation, documentation, and agent operations.
Cursor vs Codex for AI Workflows
Compare Cursor and Codex for AI coding workflows, repository assistance, debugging, refactoring, and reusable developer skills.
SKILL.md Source
# Coding-Agent Skill ## Entry Gate Only execute this skill after explicit approval in the current conversation context: - `APPROVE` Approval must correspond to the latest pending plan in-context. If no matching approved plan exists, stop and request: `Run /plan first, then reply APPROVE.` ## Execution Workflow 1. Verify toolchain before major workflows: - `command -v codex && codex --version` - `command -v timeout` - `command -v gh` - `command -v claude || test -x ~/.claude/local/claude` 2. Gather read-only context first (`rg`, `git status`, `git diff`, docs). 3. Execute the approved implementation plan. 4. Run relevant validation: - formatting/lint - typecheck - unit/integration/e2e tests as applicable 5. Report exact commands run, outcomes, and residual risk. ### Review Routing Contract (Hard Rule) - Plan artifact review must run via `scripts/plan-review` or `scripts/plan-review-live`. - PR/code review must run via `codex review --base ...` (or `safe-review.sh` wrapper). - Do not publish a manual review summary before running the review command/wrapper. ## ACP-First Routing For harness-style implementation/review requests, ACP routing is mode-specific: - Implementation uses ACP first (via `acpx`) with CLI fallback. - Review keeps review-native `codex review --base` first, then ACP fallback. - CLI fallback chain remains mandatory if ACP is unavailable. - Direct ACPX calls in coding-agent orchestration must use `scripts/acpx-direct`. - Do not emit raw `acpx ...` commands from prompts or wrappers. - Disable ACP-first per run with `CODING_AGENT_ACP_ENABLE=0`. - Override ACPX binary path with `CODING_AGENT_ACPX_CMD`. - Override ACP agent alias with `CODING_AGENT_ACP_AGENT` (default: `codex`). Known runtime limitation: - Issue #43 tracks upstream ACP spawned-run observability and relay profile alias behavior. - Use repo-side mitigation and fallback guidance in `references/acp-troubleshooting.md`. ## Verbosity Mode (Progress Updates) `CODING_AGENT_VERBOSE` controls execution progress verbosity. Default is off. - `off` (default): concise progress updates. - `on`: include structured progress updates with `Now`, `Why`, and `Next`. Accepted truthy values (case-insensitive): `1`, `true`, `on`, `yes`, `verbose`. Accepted falsy values: unset, `0`, `false`, `off`, `no`. When verbose mode is on: - Send kickoff execution status in `Now/Why/Next` format. - For long-running tasks, send periodic status updates. - Send a completion update with outcome and blockers (if any). Verbosity must not block execution. After explaining intent, proceed immediately unless waiting on a required user decision or an explicit approval gate. Status updates are not a wait state: if the next step is executable, run it in the same turn. ### Long-Run Hard-Fail Policy For wrapper/CLI runs expected to exceed 30 seconds: - Emit `RUN_EVENT start`. - Emit `RUN_EVENT heartbeat` every 20 seconds while running. - Emit `RUN_EVENT interrupted` immediately on signal/timeout/interruption. - Emit `RUN_EVENT failed` on non-interruption errors. - Emit `RUN_EVENT done` on success. ## Guardrails - No bypass-by-default. Do not use approval-bypass flags unless the user explicitly requests bypass. - Use feature branches for code changes. - Keep changes scoped to the approved plan. ## Reference Loading Open only the references needed for the current task: - `references/WORKFLOW.md` - `references/STANDARDS.md` - `references/tooling.md` - `references/codex-cli.md` - `references/claude-code.md` - `references/reviews.md`
Related Skills
coding-router
Compatibility entry skill for plan-first coding work in OpenClaw.
plan-issue
Plan-only workflow for issue/repo changes. Use when user asks to plan, scope, estimate, or design.
agentic-coding
Expert agentic coding methodologies including autonomous AI development, multi-agent systems, and self-improving code generation
typescript-coding-standards
TypeScript coding standards covering strict type system, advanced types, decorators, generics, and best practices for type-safe applications. Use for TypeScript projects requiring robust type safety and maintainable code.
swift-coding-standards
Master Swift coding standards with Apple's guidelines, protocol-oriented design, and modern concurrency patterns
rust-coding-standards
Master Rust's ownership system, type safety, and zero-cost abstractions for building safe, concurrent, and performant systems. Covers borrowing, lifetimes, traits, error handling, async/await, and testing patterns.
python-coding-standards
Python coding standards following PEP 8, type hints, testing best practices, and modern Python patterns. Use for Python projects requiring clean, maintainable, production-ready code with comprehensive testing.
kotlin-coding-standards
Master Kotlin coding standards with null safety, coroutines, and idiomatic patterns. Use when developing JVM/Android applications requiring type-safe async programming.
javascript-coding-standards
JavaScript/ES6+ coding standards following Airbnb guidelines, modern patterns, React best practices, and comprehensive Jest testing. Use for JavaScript projects requiring clean, maintainable, production-ready code with modern tooling.
go-coding-standards
Go coding standards following idiomatic Go patterns, error handling, concurrency best practices, and modern Go tooling. Use for Go projects requiring clean, efficient, production-ready code with comprehensive testing.
coding-standards
Comprehensive coding standards and best practices for maintainable, consistent software development across multiple languages and paradigms
secure-coding-practices
Secure coding practices and defensive programming patterns for building security-first applications. Use when implementing authentication, handling user input, managing sensitive data, or conducting secure code reviews.