openclaw - 💡(How to fix) Fix [Feature]: Add Cursor Agent CLI to the bundled coding-agent skill (or bless it on ClawHub) [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
openclaw/openclaw#70028Fetched 2026-04-23 07:30:11
View on GitHub
Comments
0
Participants
1
Timeline
0
Reactions
0
Participants

The bundled coding-agent skill covers Codex, Claude Code, Pi, and OpenCode, but not Cursor Agent (cursor-agent). Add Cursor as a supported engine so the bundled skill is a single canonical entry point for delegated coding work — or, if skill contributions belong on ClawHub, explicitly bless a cursor-agent skill there so downstream authors don't duplicate the pattern.

Error Message

  • macOS login-keychain gotcha (Error: Your macOS login keychain is locked. Run security unlock-keychain and try again.) and recommended mitigations.

Root Cause

The bundled coding-agent skill covers Codex, Claude Code, Pi, and OpenCode, but not Cursor Agent (cursor-agent). Add Cursor as a supported engine so the bundled skill is a single canonical entry point for delegated coding work — or, if skill contributions belong on ClawHub, explicitly bless a cursor-agent skill there so downstream authors don't duplicate the pattern.

Fix Action

Fix / Workaround

  • Update requires.anyBins["claude", "codex", "cursor-agent", "opencode", "pi"].

  • Update the description / "NOT for" line to cover Cursor.

  • Add a "Cursor CLI" section alongside "Codex CLI" / "Claude Code" / "OpenCode" / "Pi" documenting:

    • cursor-agent --print --trust --sandbox disabled 'prompt' as the default headless one-shot pattern (mirrors Claude Code's --print; no PTY required).
    • workdir:<repo> background:true pattern for long runs.
    • macOS login-keychain gotcha (Error: Your macOS login keychain is locked. Run security unlock-keychain and try again.) and recommended mitigations.
    • Review-in-temp-dir hygiene carried over from the existing Codex section.
  • Leave the "never run inside $OPENCLAW_STATE_DIR / ~/Projects/openclaw/" rule unchanged — it already applies to any coding agent.

  • Patching skills/coding-agent/SKILL.md after each npm install via a post-install hook — brittle, doesn't scale.

  • Shipping Cursor support purely as prose in per-user workspace/skills/ — fine for one user, bad for the ecosystem.

  • Running Cursor without a skill wrapper — loses the PTY / scoping / progress-update conventions and re-introduces the "agent wanders off reading $OPENCLAW_STATE_DIR" failure mode.

RAW_BUFFERClick to expand / collapse

Summary

The bundled coding-agent skill covers Codex, Claude Code, Pi, and OpenCode, but not Cursor Agent (cursor-agent). Add Cursor as a supported engine so the bundled skill is a single canonical entry point for delegated coding work — or, if skill contributions belong on ClawHub, explicitly bless a cursor-agent skill there so downstream authors don't duplicate the pattern.

Problem to solve

skills/coding-agent/SKILL.md has requires.anyBins: ["claude", "codex", "opencode", "pi"] and its body documents invocation patterns for those four CLIs only. Cursor Agent is a major, actively maintained coding CLI with a headless --print mode that fits the same shape as Claude Code (no PTY, workdir, background, same process actions for monitoring), so anyone routing work to Cursor currently has three suboptimal options:

  1. Fork the bundled skill locally — fragile, since npm install -g openclaw and openclaw doctor --fix both re-stage bundled files and will overwrite the change.
  2. Bolt on a parallel workspace skill that duplicates the coding-agent patterns just to cover Cursor.
  3. Invoke cursor-agent as a raw bash command without any of the skill-level guardrails (PTY rule, $OPENCLAW_STATE_DIR guard, review-in-temp-dir hygiene, workdir discipline, progress-update contract).

Option 2 works and ships today, but long-term it means every user running Cursor re-derives the same skill instead of consuming one canonical piece.

Proposed solution

Two options, maintainers' call:

Option A — extend the bundled skill (preferred if Cursor support is desired in core)

In skills/coding-agent/SKILL.md:

  • Update requires.anyBins["claude", "codex", "cursor-agent", "opencode", "pi"].
  • Update the description / "NOT for" line to cover Cursor.
  • Add a "Cursor CLI" section alongside "Codex CLI" / "Claude Code" / "OpenCode" / "Pi" documenting:
    • cursor-agent --print --trust --sandbox disabled 'prompt' as the default headless one-shot pattern (mirrors Claude Code's --print; no PTY required).
    • workdir:<repo> background:true pattern for long runs.
    • macOS login-keychain gotcha (Error: Your macOS login keychain is locked. Run security unlock-keychain and try again.) and recommended mitigations.
    • Review-in-temp-dir hygiene carried over from the existing Codex section.
  • Leave the "never run inside $OPENCLAW_STATE_DIR / ~/Projects/openclaw/" rule unchanged — it already applies to any coding agent.

Option B — ClawHub

If CONTRIBUTING.md's "For skill contributions, head to ClawHub" guidance means Cursor coverage belongs there, a short confirmation is all that's needed. A one-liner in skills/coding-agent/SKILL.md pointing "for Cursor, see the ClawHub cursor-agent skill" would still be very helpful so users looking at the bundled skill know where to go.

Happy to submit a PR for Option A once maintainers pick a lane.

Alternatives considered

  • Patching skills/coding-agent/SKILL.md after each npm install via a post-install hook — brittle, doesn't scale.
  • Shipping Cursor support purely as prose in per-user workspace/skills/ — fine for one user, bad for the ecosystem.
  • Running Cursor without a skill wrapper — loses the PTY / scoping / progress-update conventions and re-introduces the "agent wanders off reading $OPENCLAW_STATE_DIR" failure mode.

Impact

  • Affected: users running Cursor Agent as their primary coding CLI through OpenClaw agents. Cursor CLI has been shipping a standalone installer (~/.local/bin/cursor-agent) for several releases.
  • Severity: Low-Medium. Everything works today via user-level workspace skills; this is about removing duplicated work and giving Cursor first-class parity.
  • Frequency: Any coding task routed to Cursor — daily workload for users who prefer Cursor over Codex / Claude Code.
  • Consequence: Each user writes their own cursor-agent skill, patterns drift, and the "first-class coding CLI" list in the bundled skill becomes misleading.

Evidence / examples

  • Bundled skill (current): skills/coding-agent/SKILL.mdrequires.anyBins: ["claude", "codex", "opencode", "pi"], no mention of cursor anywhere (grep -ic cursor SKILL.md = 0).
  • Cursor CLI installs to ~/.local/bin/cursor-agent~/.local/share/cursor-agent/versions/<version>/cursor-agent on macOS.
  • Full working skill draft (matching the bundled skill's style) is ready to share on request or as a draft PR: openclaw metadata block with requires.anyBins: ["cursor-agent"], headless --print invocation, background / monitoring patterns, macOS keychain gotcha, PR-review-in-temp-dir hygiene.

Additional information

AI-assisted: this issue was drafted with help from an AI assistant and reviewed before submission. No automated PR will be opened — waiting for a maintainer signal on Option A vs Option B first.

extent analysis

TL;DR

Update the skills/coding-agent/SKILL.md file to include Cursor Agent support by adding it to requires.anyBins and documenting its invocation patterns.

Guidance

  • Review the proposed solution in the issue body, specifically Option A, which involves updating requires.anyBins to include cursor-agent and adding a "Cursor CLI" section to the skill documentation.
  • Consider the implications of Option B, which would involve creating a separate cursor-agent skill on ClawHub, and decide which approach is preferred.
  • If choosing Option A, ensure that the updated skill documentation includes the necessary information for using Cursor Agent, such as the default headless one-shot pattern and background run pattern.
  • Verify that the updated skill works as expected by testing it with Cursor Agent and ensuring that it integrates correctly with the OpenClaw ecosystem.

Example

No code snippet is provided as the issue does not require a specific code change, but rather a decision on how to proceed with adding Cursor Agent support to the coding-agent skill.

Notes

The issue highlights the need for a canonical entry point for delegated coding work and the importance of avoiding duplicated effort by users. The proposed solution aims to address this by either extending the bundled skill or creating a separate skill on ClawHub.

Recommendation

Apply workaround by extending the bundled skill (Option A) to include Cursor Agent support, as this approach provides a single canonical entry point for delegated coding work and avoids duplicated effort by users.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

openclaw - 💡(How to fix) Fix [Feature]: Add Cursor Agent CLI to the bundled coding-agent skill (or bless it on ClawHub) [1 participants]