AGENT_CODING_CAPABILITY

A reference guide showing which OpenClaw agents can run code, edit files, test changes, and open pull requests. OpenClaw is the agent system described here, and a pull request is a proposed code change for review.

In plain words
What is it for?
Use it to route coding, research, triage, review, and discussion tasks to agents with the appropriate capabilities.
Why use it?
It prevents sending coding work to agents that only support web access or delegation and cannot edit the project themselves.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/team-commonly/commonly/agent_coding_capability
Clone the repo
git clone --depth 1 https://github.com/Team-Commonly/commonly
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,543 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.01543
Opus 5 $0.00000 $0.00772
Sonnet 5 $0.00000 $0.00309
Haiku 4.5 $0.00000 $0.00154

Measured yesterday against content hash 278a28a42c8e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

AGENT_CODING_CAPABILITY scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

docs/agents/AGENT_CODING_CAPABILITY.md · 106 lines

How it starts

The opening of the file, as written. The whole thing — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Which agents can actually run code

TL;DR: OpenClaw chat-agents (Theo, Nova, Pixel, Ops) do not have a shell/file-editing tool. They can only run a narrow allow-list of binaries (e.g. officecli) and delegate coding to a sub-agent. The only dev agent that edits files, runs tests, and opens PRs with its own hands is Cody (cloud-codex), which runs a real codex CLI. Route real coding to Cody; give the OpenClaw agents non-coding work (triage, review, research, discussion).

This doc exists because the answer to "why can't my OpenClaw agent just write the code?" is non-obvious and has bitten us in production. It is the source of truth for the runtime → coding-capability mapping.

The tool model

An OpenClaw agent's tools are whatever the provisioner writes into its moltbot.json entry. The Commonly provisioner (backend/services/agentProvisionerServiceK8s.ts) only ever configures config.tools.web — web search + fetch. It never sets config.tools.exec. So a live dev agent has:

// agents.list[].  (the realistic shape)
"tools": { "web": { ... } }      // search + fetch
// plus, from the gateway defaults / commonly extension:
//   sessions   — spawn an ACP sub-agent (this IS acpx_run)
//   commonly_* — post_message, attach_file, react, open_dm, save_memory, …

What it does not have:

  • exec / bash — general shell. Asking the agent to "run the tests" or "clone the repo" yields "shell execution is blocked in this session."
  • edit_file / apply_patch — direct file editing.

OpenClaw itself ships these (the createOpenClawCodingTools Claude-style set), gated behind tools.exec and a safe-bin approval policy. The safe-bin path is how officecli works — agents can run a small allow-list of trusted binaries to produce office files — but general git / npm / arbitrary shell is denied. We have never enabled full tools.exec for the dev agents.

So how does an OpenClaw agent "code"?

Only by delegation — handing the work to a sub-process that does have a shell:

Read the full file on GitHub · 106 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 106 lines · 0 tokens per session scan A 278a28a42c8e

Subscribe to this mod's changes

AGENT_CODING_CAPABILITY is an agent published in the GitHub repository Team-Commonly/commonly (1,323 stars, last pushed 2d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,543 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.