cloud-harness-mcp: Skill for Claude Code

Cloud Harness MCP

Use Cloud Harness as a private, single-owner remote coding harness. It clones an approved repository into a TTL-limited executor and exposes bounded MCP tools. This skill guides effective and safe tool use; it does not grant credentials, host access, Docker authority, deployment authority, or permission to weaken network isolation.

Read the relevant reference

Read a reference before using an unfamiliar, destructive, networked, or recovery-sensitive operation. The references are bundled with this skill and do not require a source checkout.

Effective workflow

1. Preflight and authorization. Confirm the target repository. Use a credential-free HTTPS repository URL. Keep networkProfile at network-none unless the owner explicitly authorizes public dependency egress via dependency-access. Workspaces automatically inherit active Global Secrets for your signed-in identity. When project-specific environment credentials are also required, provide environmentId with confirmEnvironmentInjection: true (environment secrets override global secrets on key name collision). To mount agent skills or toolkits (such as mattpocock/skills, obra/superpowers, or custom Git repos), pass toolkits: [{ kind: 'preset', id: '...' }] during workspace_open.
2. Open and set active context. Call workspace_open with a fresh idempotency key. Preserve the returned opaque workspaceId exactly. Call workspace_set_active to establish default workspace context.
3. Inspect capabilities early. Run workspace_capabilities before planning write actions (e.g. git_push, github_action for issues/PRs). This prevents wasting execution effort on operations unauthorized by current GitHub App grants.
4. Inspect code with native tools. Prefer files_list, files_read, grep_search, symbols_search, and symbols_references over shell commands for code navigation. Use byte-offset limits and cursors for large files.
5. Edit with high-precision mutations.
   • Prefer files_apply_patch for single-location changes with expectedSha256.
   • Use files_write_batch for creating or updating multiple files atomically with automatic parent directory creation.
   • On CONFLICT, re-read the target file hash and rebuild the edit.
6. Execute deliberately and safely.
   • Use exec_run for synchronous, bounded single commands.
   • Use tasks_run with dependsOn for background builds, tests, or multi-step task graphs. Monitor progress via tasks_status or operation_wait.
   • Task records and output survive a runner restart (tasks_list / tasks_status); an interrupted task ends as RUNNER_RESTARTED. Interactive shell_* / sessions_* handles do not survive restart.
   • Use interactive shell_* or sessions_* only when terminal state is required.
   • For privileged: true commands in Cloudflare Access mode, expect PRIVILEGE_APPROVAL_REQUIRED and wait for the operator to approve the grant in the dashboard, then pass approvalGrantToken.
   • Injected secrets are available to container processes automatically; never attempt to print, echo, or exfiltrate secret values.
7. Git workflow and finalization.
   • Inspect status and diff using git_status and git_diff.
   • Use workspace_finalize for streamlined transactional staging, preflight diff checks, committing, and pushing to origin in a single step.
   • Pass an idempotencyKey on git_commit, git_push, and workspace_finalize. On UNKNOWN_REMOTE_STATE, retry the identical request with the same key to reconcile; treat alreadyFinalized: true as success and never re-push. git_commit expectedHeadOid is a HEAD compare-and-set returning STALE_HEAD on mismatch; git_push expectedRemoteOid (force-with-lease) returns CONFLICT with current/expected remote OIDs when the remote ref has moved.
   • Use github_action for brokered issue and pull request operations.
8. Manage lifecycle and lease. If work approaches the idle timeout, call workspace_lease_renew. If disconnected or recovering unpushed work, call workspace_recover(mode: "resume" | "status" | "patch" | "export").
9. Clean up resources. Close opened shells and sessions, cancel unfinished tasks, and always call workspace_close upon task completion, even on failure.