agentleFS
verivus-ossverivus-oss/llm-cli-gateway/.agents/skills/provider-cursor/SKILL.md

provider-cursor

Skill16 starsChanged 2 months agoMITView on GitHub

Track and maintain the upstream Cursor Agent CLI contract. Use when Cursor ships a CLI release, when a `cursor-agent` flag, execution mode, session command, ACP entrypoint, or subcommand changes, or when an upstream scan flags drift. Process guidance only: `src/upstream-contracts.ts` is the mechanical source of truth.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 3 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Keep the gateway's Cursor (cursor-agent) integration aligned with the upstream CLI.

    line 11
  2. Very specific:

    Use cursor_request for normal work and cursor_request_async for long-running work.

    line 41
  3. Distinctive:

    Keep provider-owned MCP configuration separate from gateway metadata.

    line 50
  4. Very specific:

    Use a real resumable Cursor chat only when the response reports it as resumable.

    line 68
  5. Very specific:

    Follow multi-llm-review's full-access protocol, inspect the current request schema and live capability response, and use a fresh …

    line 86
  6. Distinctive:

    Add conformance fixtures that prove every new accept and reject behavior.

    line 112
  7. Very specific:

    If a source URL or watch category changes, update TypeScript metadata first, then mirror it in docs/upstream/provider-sources.dag.toml.

    line 113
  8. Very specific:

    Verify with npm run build, npm run upstream:contracts, and targeted provider tests before the full suite.

    line 115
  9. Distinctive:

    Keep destructive or interactive commands, including worker startup and any session mutation, catalog-only unless an approval-gated product …

    line 124

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Provider: Cursor Agent CLI

Keep the gateway's Cursor (cursor-agent) integration aligned with the upstream CLI. This skill does not define argv or environment behaviour. The single mechanical source of truth is UPSTREAM_CLI_CONTRACTS.cursor in src/upstream-contracts.ts, enforced by validateUpstreamCliArgs and validateUpstreamCliEnv. Do not re-encode flags, execution modes, ACP rules, or session rules here or in TOML.

Identity

FieldValue
CliTypecursor
Executablecursor-agent
DistributionCursor Agent vendor distribution
CLI docshttps://cursor.com/cli
ACP docshttps://cursor.com/docs/cli/acp
Parameter referencehttps://cursor.com/docs/cli/reference/parameters
Watch categoriesflags, subcommands, session-resume, execution-modes, acp-entrypoint

These values mirror UPSTREAM_CLI_CONTRACTS.cursor.upstreamMetadata and docs/upstream/provider-sources.dag.toml ([providers.cursor]). The TypeScript metadata is authoritative; the TOML is scanner input only.

Gateway use

  1. Discover the live gateway surface before relying on Cursor-specific controls:
    provider_tool_capabilities({cli:"cursor"})
    For a cached read-only resource, use provider-tools://cursor.
  2. Use cursor_request for normal work and cursor_request_async for long-running work. Sync calls may auto-defer; poll llm_job_status and fetch with llm_job_result when that happens.
  3. Omit model unless the caller explicitly requests one. Use only controls reported by provider_tool_capabilities, especially execution mode, sandbox, workspace, and session controls. Cursor must use approvalStrategy:"legacy": it rejects mcp_managed before launch because ambient MCP configuration cannot be isolated, and approvalPolicy has no effect.
  4. Keep provider-owned MCP configuration separate from gateway metadata. Do not assume every Cursor CLI control is safe to expose through a request.
  5. Cursor accepts flat prompt only, not promptParts. workingDir selects the child process cwd on transport:"cli", as it does for every other provider. It is CLI-only: transport:"acp" rejects it in rejectUnsupportedCursorAcpParams alongside addDir, because the ACP route resolves its own scope and would otherwise discard it silently. Its workspace field is separate and selects a local target directory, registered alias, or provider-native .code-workspace file. A directory/alias establishes child cwd; a workspace file remains a native argument and is not misused as spawn.cwd. Because both inputs can name a cwd, resolveCursorWorkingDir rejects an absolute workspace path that disagrees with workingDir instead of ranking them; identical values are admitted. Verify the target before reviewing concurrent repositories. An unregistered relative value remains a provider-native saved-workspace name and is passed through verbatim, never resolved against the gateway process cwd. Use an absolute path when selecting an unregistered local directory as child cwd.
  6. Use a real resumable Cursor chat only when the response reports it as resumable. Gateway bookkeeping IDs are not automatically valid Cursor chat IDs. Continuation remains provider-native under legacy. A cwd-scoped resumeLatest needs a stable selected target.
  7. Native ACP is capability-gated. It rejects approvalStrategy:"mcp_managed" and any approvalPolicy, as does the Cursor CLI gateway path. Probe cursor-agent acp --help when validating the installed entrypoint.
  8. An unscoped CLI child uses a fresh neutral temporary cwd, not the gateway repository. Cursor's argv-bound prompt rejects oversized UTF-8 input as non-retryable input_too_large; the gateway never truncates it. All other caller-controlled argv values are admitted in their final encoded form before spawn. Embedded NUL bytes return non-retryable invalid_input without exposing the rejected value.

Full-access review handoff

For a user-authorized exhaustive review that needs full provider permissions and native MCP access, do not infer the launch controls from this maintenance guide. Follow multi-llm-review's full-access protocol, inspect the current request schema and live capability response, and use a fresh target-checkout node dist/index.js --transport=stdio gateway rather than a global process. Reapply the grant on every new job, preserve native MCP configuration without a pretend gateway allowlist, send the corrective-program report and exact diff/file identity, set no caller caps, and honor a user-required 90-second progress cadence. The reviewer must independently inspect code, docs, tests, and commands before it can approve.

Scan for upstream change

npm run upstream:scan -- --provider cursor
npm run upstream:scan -- --live --provider cursor --fail-on-critical
npm run upstream:scan -- --provider cursor --probe-installed --fail-on-critical
npm run upstream:scan -- --live --provider cursor --probe-installed --write-snapshot --write-report

A failed source fetch is advisory unless --fail-on-critical is passed. Use the installed help probe as the primary signal for rapidly changing command surface.

Update procedure

  1. Edit UPSTREAM_CLI_CONTRACTS.cursor in src/upstream-contracts.ts: flags, arities, enums, execution modes, session rules, ACP probe commands, and catalogued subcommands belong there.
  2. Add conformance fixtures that prove every new accept and reject behavior.
  3. If a source URL or watch category changes, update TypeScript metadata first, then mirror it in docs/upstream/provider-sources.dag.toml.
  4. Verify with npm run build, npm run upstream:contracts, and targeted provider tests before the full suite.

Cursor-specific notes

  • Tested against cursor-agent 2026.09.02-c22c1a3.
  • Root commands include authentication, MCP, worker, update, chat/session, and rule-management surfaces. Catalog them with an explicit risk and exposure tier before exposing any of them through gateway admin or request tools.
  • Keep destructive or interactive commands, including worker startup and any session mutation, catalog-only unless an approval-gated product surface is deliberately added.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.